A series of security vulnerabilities have been disclosed in internet infrastructure software within a single day, affecting PHP, Node.js, and Go.
Multiple internet infrastructure software projects rolled out security updates in a single day, addressing vulnerabilities including remote code execution, TLS certificate validation bypass, and certificate trust risks. PHP fixed two critical flaws: CVE-2026-109618 and CVE-2026-109619, both enabling unauthenticated remote code execution that impacts PHP environments processing web requests. The patched versions are 8.5.12, 8.4.27, 8.3.36, and 8.2.35. Node.js’ advisory details 10 vulnerabilities, four of which are high-severity: these include remote code execution via out-of-bounds write in the Brotli decoder, DNS name truncation, and two TLS certificate hostname validation bypass issues. Fixed versions are 26.12.1, 24.22.1, and 22.23.4. Additionally, Go’s crypto/x509-related risk is tracked as CVE-2026-111943, rated "Important" by Red Hat with a CVSS 3.1 score of 6.8. Attackers with specific network access and valid certificates could impersonate legitimate servers to intercept or alter communications. Go 1.26.10 and 1.27.3 have been added to a deny list, blocking trust for compromised intermediate certificates.
53 minutes ago
Dataline completes $10.6 million seed round funding, and plans to conduct an ICO.
Decentralized data network for AI agents Dataline has closed a $10.6 million seed round, co-led by Lattice and Samsung-backed Dispersion Capital, with participation from Symbolic Capital, Coinbase Ventures, Anagram, and Robinhood co-founder & CEO Vlad Tenev, among others. The round, completed in April, utilized a SAFE agreement with token warrants; valuation was not disclosed. Dataline is now live, covering approximately 50 data sources to deliver on-chain and off-chain data, and scores data reliability, timeliness, and consistency using AI. Proceeds will be allocated to expanding data sources, enhancing data validation, and expanding into robotics, AI training data, and other sectors. Dataline plans to launch a token by the end of 2026 for network payments and data access, and intends to conduct a public token sale on Base via Uniswap’s Continuous Clearing Auction (CCA) system, with details set to be revealed next week. The public sale’s target fundraising amount, token quantity, and other terms remain undisclosed.
53 minutes ago
Trump plans to leverage White House cyber crime efforts to showcase achievements in combating cross-border crypto fraud.
White House officials said on October 21 that Trump planned to focus on the U.S. government’s actions to crack down on overseas crypto fraud rings during the White House’s cybersecurity event that day, including recovering funds and shutting down fraudulent operations. These transnational criminal groups steal funds from U.S. citizens via romance scams, phishing, ransomware and other methods. The U.S. government noted that law enforcement agencies have seized approximately $833 million in crypto assets linked to overseas fraud, and shut down 503 fraudulent crypto websites under the supervision of the Homeland Security Task Force. Relevant authorities have also notified nearly 9,000 U.S. citizens to alert them they are being targeted by fraud. The U.S. Treasury Department previously announced the establishment of an inter-agency "Fraud Strike Task Force" to target "pig butchering" scams run by organized criminal groups in Myanmar, Cambodia, Laos and the Philippines. Since last year, the U.S. has also imposed multiple sanctions aimed at cutting off support for related fraud websites and their infrastructure.
53 minutes ago
American Bankers Association voices dissatisfaction with OCC’s final GENIUS Act rule, which abandons proposed restrictions on stablecoin yields.
Crypto journalist Eleanor Terrett reported that the American Bankers Association (ABA) has voiced dissatisfaction with the final rules implemented by the U.S. Office of the Comptroller of the Currency (OCC) under the GENIUS Act. The OCC abandoned the previously proposed restrictions on stablecoin yields, a move the ABA argues will "maintain the status quo for crypto platforms." The ABA’s CEO stated: "It is a mistake to wait until deposit outflows and economic harm have already occurred before attempting to address these issues."
53 minutes ago
Beyond PHP, three Node.js version branches are also affected by high-severity vulnerabilities, and users need to update their runtime environments promptly.
Node.js rolled out a security update on October 21, covering its 26.x, 24.x, and 22.x version branches, with corresponding patched releases: 26.12.1, 24.22.1, and 22.23.4. The advisory details 10 total vulnerabilities, categorized as 4 critical-severity, 3 medium-severity, and 3 low-severity, with varying impact scopes across the three version lines. The critical CVE-2026-111497 is an out-of-bounds write flaw in the Brotli decoder that could enable remote code execution (RCE), affecting all three branches. Other critical issues include improper null character handling in DNS resolution and a TLS certificate hostname validation bypass. The update also resolves problems with persistent connection response processing, random number cache duplication, and permission models, while updating dependencies including c-ares, OpenSSL, and Undici. The official note warns that end-of-life (EOL) versions may still be vulnerable, advising users to use the latest supported releases. Separately, prior reports noted PHP has fixed two critical unauthenticated RCE vulnerabilities, with patches available for versions 8.5.12, 8.4.27, 8.3.36, and 8.2.35; PHP users are urged to upgrade immediately.
53 minutes ago
Bonk Guy: This round is the "privacy cycle", privacy protocol MASK is undervalued.
Renowned trader Bonk Guy posted that the current crypto market is showing characteristics of a "privacy cycle". He noted that since yesterday, Bitcoin has risen approximately 3%, ZEC is up 16%, adding over $3 billion in market capitalization. MASK currently has a market cap of roughly $26 million, with its protocol trading volume and fees hitting consecutive all-time highs, he added. Citing data from DefiLlama, he stated that MASK ranks first among privacy protocols by revenue, outearning the next 10 privacy protocols combined. By comparison, ZEC’s market cap gain over the past 24 hours is roughly 115 times MASK’s entire market cap. Bonk Guy also said that MASK’s upcoming deployment on Solana is set to launch in the coming days, which he expects will further drive growth in usage, trading volume, and fees. He went on to call MASK the most undervalued privacy protocol in the crypto market.
Not financial or tax advice. PANews content is strictly educational and informational and is not investment advice, financial advice, tax advice, legal advice, or a solicitation to buy or sell any digital asset, security, or financial product. Do your own research and consult qualified advisers.
Disclosure. PANews may publish sponsored content, partner content, advertisements, affiliate links, event promotions, and market commentary involving Web3 projects, service providers, or financial products. PANews personnel, contributors, or affiliates may hold digital assets or other interests related to covered topics. See our Terms of Service.
The Karpelès Lab CEO says he found the implant in a Ledger Flex and a Nano X and that its newest version is built for mass production, while Ledger says German store stock it inspected showed no tampering.
Mark Karpelès says the Ledger implant turned up in devices from two more authorized resellers
Posted October 16, 2026 at 1:55 pm ET.
Mark Karpelès, CEO of Tokyo-based Karpelès Lab, said the spy implant found in Ledger hardware wallets sold by Southeast Asian reseller CryptoBilis has also turned up in devices from two more authorized Ledger resellers, and that its newest version is built for mass production.
Karpelès said on Unchained’s Oct. 15 livestream that one of those devices was a Ledger Flex bought from Metapod, a reseller with stores at Singapore’s Changi Airport. He said the implant was customized for the Flex, but “it’s still the same kind of board and the same process of having three chips.” He said one chip reads the screen, one is a cellular modem and one is an eSIM that sends the data.
The other, he said, was a Nano X bought at MediaMarkt in Germany by X user Johannes, who Karpelès said sent one unit to him and one to Ledger. Karpelès said “the one sent to me arrived today” and that he found an implant inside, installed a little differently from the Southeast Asian units. “It’s surely the same implant,” he said.
Karpelès first reported the Flex on X on Oct. 13. Johannes shared photos of his MediaMarkt unit on Oct. 11, and Karpelès posted his own MediaMarkt result earlier on Oct. 15.
Get Unchained’s crypto news in your inbox with the free Unchained Daily newsletter.
What Ledger Has Said
Ledger has said every confirmed drain from the Oct. 9 wave involves CryptoBilis devices. MediaMarkt Saturn halted Ledger sales on Oct. 13 “with Ledger’s alignment,” and Ledger asked sellers in seven Asian markets, including Singapore, to pause sales.
Ledger said in an Oct. 14 post that its team inspected devices from retail and distributor stock in Germany, including the stores named, and “none showed any sign of tampering.” It added, “We have not yet been able to examine the devices shown online.” Johannes replied that the store’s other units appeared to be clean.
Karpelès said he “wouldn’t be able to say for sure” whether Johannes found MediaMarkt’s only tampered units.
‘Made for Mass Production’
In Karpelès’s telling, the implant has gone from a board wired into the device to blank boards that clip in, and on the Nano X “the board will basically just fall into place.” “They spent months of research and development building this,” he said, citing “a lot of funding” behind it.
Implants like this were once costly and narrowly targeted, Karpelès said, but he described the new one as made “for mass production” and said “we’re likely going to see a lot more of it” in the future.
Asked whether Ledger should end its authorized reseller program, Karpelès said that would be “probably too drastic.” He suggested tamper-evident packaging.
Related Listen: The Chopping Block: The Ledger Hardware Implant, a Listening Device at OpenAI, Papertrade’s 1000x House, and the Death of Technical Moats
AI-assisted content: This article was produced with the assistance of AI tools and was reviewed, edited, and fact-checked by a member of the Unchained editorial team before publication.
Hong Kong Exchanges and Clearing (HKEX) aims to transition to T+1 settlement in the second half of 2028.
Hong Kong Exchanges and Clearing (HKEX) has published a consultation summary document on shortening the settlement cycle of Hong Kong’s spot stock market, which shows widespread market support for implementing a T+1 settlement cycle in Hong Kong’s securities market. Based on feedback from various stakeholders, HKEX plans to complete the relevant transition in the second half of 2028, provided the market is ready and regulatory approval is obtained. The consultation received a total of 109 non-duplicate responses from market stakeholders including brokers, custodians, investment managers, industry associations, technology vendors, and investors. Respondents supported Hong Kong’s adoption of a T+1 settlement cycle, agreeing that this move will help enhance market efficiency, reduce settlement risks, improve capital efficiency, and consolidate Hong Kong’s competitiveness as an international financial center.
1 seconds ago
Microsoft completes pre-training of a 2.8-trillion-parameter model, outperforming the GPT-5.6 base model with only one-quarter of the computing power.
Beating AI News: Microsoft announces completion of pre-training for its new self-developed model MAI-Base-2. The model has a total of 2.8 trillion parameters, adopts a Mixture-of-Experts (MoE) architecture, and activates only 80 billion parameters per forward pass. Between August and September this year, Microsoft used 16,000 NVIDIA GB200 GPUs for 38 days of training, processing approximately 50 trillion tokens. The new model also introduces native visual pre-training for the first time, with visual data accounting for 18% of its training data. Internal tests released by Microsoft show that MAI-Base-2 outperforms models including DeepSeek-V4-Pro-Base, Kimi K3, and Qwen3.8 on text prediction benchmarks for code, question answering, mathematics, and science and engineering disciplines. Microsoft also claims that training a model of equivalent performance to DeepSeek-V4-Pro-Base would require only 1/4 to 1/12 of the computing power using its approach. However, Microsoft has invested additional computing power to build a larger, more capable model; MAI-Base-2’s actual training computing power consumption is still higher than that of DeepSeek. Microsoft AI head Mustafa Suleyman stated at an internal meeting on October 15 that MAI-Base-2’s pre-training base is already stronger than the OpenAI model supporting GPT-5.6 Sol, while requiring only one-fourth the training computing power. Microsoft improved MoE expert scheduling, Muon optimizer, and data filtering methods to reduce GPU memory waste and enhance the stability of large-scale training. The company also emphasized that all training data is processed in-house, and the model’s capabilities were not obtained through distillation of other firms’ models. MAI-Base-2 is currently in the post-training phase. Microsoft plans to first deploy it for its own products, then license the model weights to enterprise partners via Foundry.
1 seconds ago
Lily Liu responds to the Solana Shanghai event: Overwhelming sign-ups have led to excess attendance, so the event will be postponed to make adequate preparations.
Solana Foundation Chair Lily Liu posted that the large-scale light display on the entire exterior wall of Citigroup Building in Shanghai’s Lujiazui area on the evening of October 14 was meant to showcase Solana’s deep affection for the Chinese community and express gratitude for the warm welcome extended by the community. Due to promotion efforts, registrations for the "Solana Shanghai Event" quadrupled overnight, far exceeding the maximum number declared to local authorities. Officials were unable to prevent the excessive influx of attendees. To comply with local regulations, the event will be postponed, pending a time when it can better accommodate the community’s scale and reach. We extend our apologies to all friends who traveled near and far to join the Shanghai Solana event.
1 seconds ago
Lightning Labs unveils LayerTree, a Bitcoin scaling solution that allows up to around 1 million users to share a single UTXO.
Lightning Labs has proposed a Bitcoin scaling solution called LayerTree, a non-custodial architecture that allows up to around 1 million users to share a single UTXO and connect to the Lightning Network. Balances are recorded in an off-chain Merkle-sum tree, with the shared output committing to the tree’s root, enabling users to exit unilaterally via a single transaction. The scheme depends on as-yet-unactivated Bitcoin consensus layer changes—such as covenant-related opcodes—and cannot be directly deployed on the mainnet at this time. The project team states that LayerTree is a design and reference implementation, and the opcodes it uses remain proposals rather than active consensus rules.
1 seconds ago
Vercel Adds Background Persistence to Its Agents: Tasks Continue Processing, Files Are Not Lost Even When Sandboxes Are Destroyed
Beating AI Express News: Vercel launched Daemons at its Ship 2026 Developer Conference, enabling developers to directly deploy long-running background processes. These processes can continuously pull tasks from message queues or databases, process one after another, and automatically restart if they crash. Previously, Vercel primarily executed code on a request basis, so developers needing to run long-running tasks often had to deploy separate background services. Daemons is now in public beta. Another update is the official launch of Drives, which provides persistent file storage for AI Agents. Previously, files inside temporary sandboxes would disappear once the sandbox was shut down. Now, Agents can save code, data, and work products in independent storage for reuse when a new sandbox is launched next time. Developers can also duplicate workspaces, allowing multiple Agents to start from the same set of files, process different tasks in parallel, and avoid overwriting each other’s modifications. Vercel also added MCP Events to Connect, enabling Agents to subscribe to events from external systems. For example, when a new GitHub Issue is created or a document receives a new comment, compatible MCP services can proactively send notifications, triggering Agents to process them without repeated polling for new tasks. This feature is also in public beta and requires external services to support event subscriptions. These updates expand Vercel’s business scope. Developers who previously used the platform mainly for website deployment can now host long-running background tasks, Agent workspaces, and event integrations all on Vercel, reducing the need to build and maintain separate infrastructure.
1 seconds ago
UMX officially launched today, supporting Hong Kong and US stock trading, and rolling out the industry’s first crypto-collateralized stock trading service.
Crypto-friendly securities platform UMX, incubated by Avenir Group, announced its official launch and full open registration. Since the public beta kicked off on August 10, UMX has collected 788 product feedbacks, with 89 demands incorporated into the official version, completing a practical validation round rooted in real use scenarios and user needs. It is reported that UMX stems from the cross-asset management practice of Li Lin’s family office team. To break down barriers between crypto assets and securities, UMX has built a unified trading system covering all categories of Hong Kong and U.S. stocks as well as crypto assets. The platform adopts cross-asset unified margin, flexible transfers, and the industry-first "hold-assets-for-credit" stock purchasing mechanism, enabling users to directly access purchasing power for Hong Kong and U.S. stocks without selling their crypto assets, realizing the function of "buying stocks via credit based on held assets". During the launch period, UMX is running a "Grand Opening Triple Rewards" campaign: users who meet wealth management targets will receive the latest iPhone Duo, earn UMX points for trading activities, and new users can get up to 100 USDT in bonus rewards.
Crypto users are reaching for scales after a tampered hardware wallet scandal tied to roughly $93 million in reported thefts. However, this so-called proof of weight can expose an implant without proving a device is clean.
The check gained traction after Ledger confirmed a hidden implant in a device sold through CryptoBilis, a reseller Ledger lists for Southeast Asia. Ledger says its own systems remain uncompromised, yet the affected device reportedly reached its buyer through that listed channel.
Why Owners Are Putting Their Hardware Wallet on a ScaleWeighing appeals because it leaves a sealed device untouched. Manufacturers publish reference figures for comparison.
Ledger lists the Nano X at 34 grams, while Trezor puts its Safe 3 at 14 grams. Owners then compare that figure with their own reading.
In theory, an extra circuit board, antenna, and cellular modem add mass, so a heavier hardware wallet looks suspicious.
Former Mt. Gox CEO Mark Karpelès found that setup in a sealed Nano X bought in Malaysia. According to him, a hidden spy chip read the recovery phrase off the screen and sent it out over a cellular connection.
Why a Few Grams Prove Little Either WayHowever, the implant never needed a perfect match. In Karpelès’ unit, the board sat behind the screen while a shorter battery freed room, shrinking the added weight.
Karpelès himself put the remaining difference at roughly one gram. According to him, that falls within normal factory variance, since batteries of equal capacity can differ in weight.
The difference varies by ~1g, which might be just what your ledger weight may vary due to factory variables (batteries for example tend to not weight the same even for equal capacity).
— Mark Karpelès (@MagicalTux) October 11, 2026 A kitchen scale reading in one-gram steps cannot separate that gap from a small implant. Precision micro scales do better, but only when calibrated against a reference weight. Opening the casing remains the only direct way to inspect a hardware wallet board.
Ledger’s own advice goes further. The company paused reseller sales and urged recent CryptoBilis buyers to skip hardware wallet setup entirely.
Ultimately, a kitchen scale only catches a crude implant, not a careful one. As hardware wallet attacks move from software into the device itself, buying straight from the manufacturer may matter more than any home test.
Not financial or tax advice. PANews content is strictly educational and informational and is not investment advice, financial advice, tax advice, legal advice, or a solicitation to buy or sell any digital asset, security, or financial product. Do your own research and consult qualified advisers.
Disclosure. PANews may publish sponsored content, partner content, advertisements, affiliate links, event promotions, and market commentary involving Web3 projects, service providers, or financial products. PANews personnel, contributors, or affiliates may hold digital assets or other interests related to covered topics. See our Terms of Service.
Analysis: Bitcoin remains in the early stage of a long-term bull market, and may continue to consolidate sideways in the short term.
Crypto analyst Willy Woo said in a post that the Puell Multiple indicator currently shows Bitcoin (BTC) remains in the early stages of a bull market from a long-term structural perspective, while the market is in a sideways consolidation phase from a near-term structural standpoint. The Puell Multiple is used to roughly gauge the selling pressure exerted by Bitcoin miners.
3 minutes ago
Wells Fargo is in discussions with Payward, the parent company of Kraken, over a potential partnership that would help it expand its cryptocurrency trading operations.
U.S. bank Wells Fargo is in discussions with Payward, the parent company of crypto exchange Kraken, for a potential collaboration under which Payward would provide liquidity support for the bank’s digital asset trading business. The partnership would see Payward offer market depth and trade execution support, helping Wells Fargo expand crypto asset trading access for its clients without building out a full trading infrastructure from scratch. The two sides are still in negotiations, no final agreement has been reached, and both parties declined to comment. Wells Fargo has previously offered Bitcoin spot ETFs to eligible wealth management clients, invested in compliance tech firm Elliptic and trading platform provider Talos, advanced a blockchain-based deposit initiative, and joined a U.S. dollar stablecoin-related consortium. In September 2026, Nasdaq announced a $100 million investment in Payward, valuing the firm at roughly $21 billion, with Wells Fargo serving as the exclusive capital markets advisor for the deal. Separately, Payward is reportedly in talks with asset custody and services giant BNY for a broader infrastructure partnership, covering areas including digital asset products, custody, wealth management, trading, and payments. Those discussions have also not yet resulted in a final agreement. Payward is steadily expanding its footprint in traditional finance. The firm is also reportedly pursuing a European banking license, with Lithuania as its target jurisdiction. In March 2026, Kraken Financial became the first digital asset bank to secure access to Federal Reserve payment infrastructure; Payward has also obtained authorization from the Dubai Virtual Assets Regulatory Authority.
3 minutes ago
A certain whale took a 10x short position on ENA, with returns reaching 560%.
According to TradingBeats monitoring, a Hyperliquid whale address 0x45d2…4029 holds approximately 6.416 million ENA short positions, with a position value of around $1.54 million, using 10x full leverage at an average entry price of roughly $0.552. The address is currently sitting on an unrealized profit of about $2 million, translating to a position return of approximately 564.5% based on the initial leverage. This address has held ENA short positions since late September. As of press time, the ENA contract’s mark price is around $0.24, some 56.5% lower than the address’s average entry price.
3 minutes ago
Former Acting Chair of the CFTC: No congressional legislation is needed, as the existing framework is sufficient to regulate the crypto market.
Former acting Chair of the U.S. Commodity Futures Trading Commission (CFTC) and current Chief Legal Officer and Chief Administrative Officer at MoonPay, Caroline Pham, stated that the Commodity Exchange Act grants the CFTC broad regulatory authority, meaning the existing regulatory framework can cover emerging commodity sectors like crypto assets without waiting for congressional legislation. Pham believes the CFTC’s recently proposed rules will help strengthen protections for U.S. retail crypto traders. She noted that integrating relevant transactions into strictly regulated trading platforms is more conducive to investor protection than separate regulation by individual U.S. state laws. On stablecoins and tokenized deposits, Pham said tokenized deposits are more suited for wholesale financial markets. If all trades in a single trading day were settled via real-time atomic settlement on a blockchain, the required cash volume could hit $15 trillion—an amount existing funds cannot support. Banks currently offset counterparties’ receivables and payables through net settlement, so tokenized deposits will play a key role in wholesale markets; stablecoins, by contrast, are better for scenarios where counterparties lack trust and need to mitigate settlement risk. Additionally, Pham projects that over the next two years, the New York Stock Exchange (NYSE) and Nasdaq will begin listing genuine tokenized stocks and generate substantial trading volume. She stressed that real tokenized stocks should retain shareholder rights such as voting power, distinguishing them from synthetic tokens not approved by the issuing company.
3 minutes ago
Fudan University's Qiu Xipeng team raises over RMB 1 billion in funding, plans to build an AI assistant similar to the one in the movie *Her*
Beating AI News (Dongcha) — Mosi Intelligence, founded by Fudan University professor Qiu Xipeng, announced the completion of over RMB 1 billion in Pre-A round financing, led by Huake Fund, with participation from Guokai Science and Technology Innovation and Shanghai Science and Technology Innovation Fund, among others. Founded at the end of 2024, the company is led by Li Shimin, a student of Qiu Xipeng, as its CEO. The proceeds from this round will be used for research and development of full-modal real-time interactive technology and global commercialization. Mosi aims to build AI that not only understands human speech but also grasps users' emotions and surrounding context. For example, if a user says "It's a bit cold" in a car, the AI will adjust the air conditioning based on in-vehicle temperature, the speaker's identity, and past preferences. It also needs to know when to respond and when to remain silent. Technically, the team has open-sourced models including MOSS-Audio and MOSS-VL-Realtime. The former can analyze emotions in speech, identify different speakers, and recognize ambient sounds. The latter can continuously process video, answer questions in real time while viewing, and revise responses based on changes in the visual feed. Currently, Mosi is integrating voice, vision, memory, and tool calling capabilities into the MOSS Live real-time interaction system, targeting use cases such as smart cockpits and AI hardware. The company said it has already delivered products to some enterprise clients, with a long-term goal of building an AI assistant like the one in the movie *Her*, capable of continuously understanding the user's context.
3 minutes ago
Hong Kong, China will list its second spot Solana ETF on October 13.
According to official announcements from Solana, Hong Kong will launch its second spot Solana ETF. The Pando Solana ETF (2831.HK) is scheduled to list on the Hong Kong Stock Exchange on October 13. Solana noted that the total net asset value of global spot SOL ETFs currently stands at $1.9 billion, with cumulative net inflows reaching $1.61 billion.
Insight Beating AI Flash News: Google Labs has launched Playground, an AI-powered game platform that allows users to generate playable web games simply by describing their ideas in text—no programming experience required. Users can create games from scratch, or select genres like trivia, tower defense, racing, etc., while specifying 2D or 3D, single-player or multiplayer modes. After generation, users can modify game rules, characters, scenes, and physics effects via chat. The platform also supports uploading images for AI to convert into game assets. Completed games run on mobile and desktop browsers, can be shared via links, or published to the community for others to play. Some game types support multiplayer online play and leaderboards. Playground leverages models including Gemini, Nano Banana, and Lyria. Free users can generate games but face weekly usage limits, while eligible Google One AI subscribers get higher quotas. Google also announced a collaboration with Unity, planning to integrate Unity Spark to enable creation of more complex, higher-fidelity 3D games. However, Spark remains in testing and is not yet officially launched. Playground is currently only available to users aged 18 and above in the United States.
In brief Google released Nano Banana 2.1 on Oct. 6, rolling it out in the Gemini app, Search's AI Mode, Google Ads and its developer tools. Google's own tests put it ahead of Nano Banana 2 and Nano Banana Pro, including an infographic factuality score of 0.521 versus 0.179, but independent testing was not available at launch. Developer pricing is about half of Nano Banana 2's, starting at $0.0336 for a standard 1K image. Google on Tuesday released Nano Banana 2.1, the newest version of the AI model that turns a typed sentence into a picture, or edits a photo you already have.
It's rolling out in the Gemini app, Google Search's AI Mode, Google Ads and developer tools including Google AI Studio, Flow and Stitch.
Myriad: Which company will IPO next? Click to make your prediction.The original Nano Banana made Gemini a hit. In September 2025, its trick of turning selfies into collectible figurines pushed Gemini to No. 1 on both major app stores, dethroning ChatGPT after nearly three years.
Alphabet's market value topped $3 trillion in the same stretch.
Version 2 arrived in February, built on Gemini 3.1 Flash Image. Its big trick was checking Google Search before drawing real-world things, so a picture of an actual event or person had a better shot at being right.
Meet Nano Banana 2.1, our latest image generation and editing model 🍌
This upgraded version outperforms our previous models across the board, with notable leaps in visual design, mask-based editing, and subject consistency to help you create more natural-looking images. pic.twitter.com/d858DrzrF3
— Google (@Google) October 6, 2026
So what's new in this update? Google points to three upgrades: better visual design, sharper mask-based editing (you mark one region and only that area changes), and better subject consistency, meaning a person or object still looks like itself after several edits.
On overall preference in text-to-image tests, Nano Banana 2.1 scored 1,050 ELO Points. Nano Banana 2 scored 990 and Nano Banana Pro scored 935. In the ELO system, the more points the better. It doesn’t have a ceiling, so it is purely qualitative.
On paper, it handles up to 14 reference images at once, tracking up to four characters and ten objects. It outputs up to 4K, roughly TV-level sharpness, in formats as wide as 8:1.
Developers can also set how long it "thinks" before drawing, from minimal to high, and let it ground its results in Google Search and Google Image Search. In plain words, it can look things up before it draws them.
Through Google's developer API, the pipe that lets other companies' apps use the model, a standard 1K image costs $0.0336. That's about half of Nano Banana 2's $0.067.
A 4K image runs $0.0756 versus $0.151. Batch jobs, big piles of images processed in bulk instead of instantly, get another 50% off.
At the standard 1K price, a thousand images now cost about $33.60 through the API, down from $67 on Nano Banana 2, which makes this model extremely good and also very efficient and cost effective.
Daily Debrief NewsletterStart every day with the top news stories right now, plus original features, a podcast, videos and more.
In brief Google released Nano Banana 2.1 on Oct. 6, rolling it out in the Gemini app, Search's AI Mode, Google Ads and its developer tools. Google's own tests put it ahead of Nano Banana 2 and Nano Banana Pro, including an infographic factuality score of 0.521 versus 0.179, but independent testing was not available at launch. Developer pricing is about half of Nano Banana 2's, starting at $0.0336 for a standard 1K image. Google on Tuesday released Nano Banana 2.1, the newest version of the AI model that turns a typed sentence into a picture, or edits a photo you already have.
It's rolling out in the Gemini app, Google Search's AI Mode, Google Ads and developer tools including Google AI Studio, Flow and Stitch.
Myriad: Which company will IPO next? Click to make your prediction.The original Nano Banana made Gemini a hit. In September 2025, its trick of turning selfies into collectible figurines pushed Gemini to No. 1 on both major app stores, dethroning ChatGPT after nearly three years.
Alphabet's market value topped $3 trillion in the same stretch.
Version 2 arrived in February, built on Gemini 3.1 Flash Image. Its big trick was checking Google Search before drawing real-world things, so a picture of an actual event or person had a better shot at being right.
Meet Nano Banana 2.1, our latest image generation and editing model 🍌
This upgraded version outperforms our previous models across the board, with notable leaps in visual design, mask-based editing, and subject consistency to help you create more natural-looking images. pic.twitter.com/d858DrzrF3
— Google (@Google) October 6, 2026
So what's new in this update? Google points to three upgrades: better visual design, sharper mask-based editing (you mark one region and only that area changes), and better subject consistency, meaning a person or object still looks like itself after several edits.
On overall preference in text-to-image tests, Nano Banana 2.1 scored 1,050 ELO Points. Nano Banana 2 scored 990 and Nano Banana Pro scored 935. In the ELO system, the more points the better. It doesn’t have a ceiling, so it is purely qualitative.
On paper, it handles up to 14 reference images at once, tracking up to four characters and ten objects. It outputs up to 4K, roughly TV-level sharpness, in formats as wide as 8:1.
Developers can also set how long it "thinks" before drawing, from minimal to high, and let it ground its results in Google Search and Google Image Search. In plain words, it can look things up before it draws them.
Through Google's developer API, the pipe that lets other companies' apps use the model, a standard 1K image costs $0.0336. That's about half of Nano Banana 2's $0.067.
A 4K image runs $0.0756 versus $0.151. Batch jobs, big piles of images processed in bulk instead of instantly, get another 50% off.
At the standard 1K price, a thousand images now cost about $33.60 through the API, down from $67 on Nano Banana 2, which makes this model extremely good and also very efficient and cost effective.
Daily Debrief NewsletterStart every day with the top news stories right now, plus original features, a podcast, videos and more.
Beating AI News Flash: Google officially launched Nano Banana 2.1, an upgraded iteration of Nano Banana 2 that remains positioned as a high-efficiency image generation and editing model. Key enhancements span image quality, prompt adherence, text generation, and multi-turn character consistency. Support for the model is now live across Google products including Gemini, AI Mode, AI Studio, Flow, Stitch, and Google Ads, with its API identifier set as gemini-nano-banana-2.1. Image editing is a core focus of this update: Version 2.1 supports 1K, 2K, and 4K resolutions, allows up to 14 reference images at once, and maintains consistency for up to 4 characters and 10 objects. Google has also improved mask editing (modifying only specified areas), plus enhanced infographic layout, image text, and ultra-wide aspect ratio generation. The model now offers three thinking intensity tiers: minimal, medium, and high. In Google’s internal model benchmarks, the thinking-enabled Nano Banana 2.1 outperforms Nano Banana 2 and Nano Banana Pro across all 10 evaluated metrics, including text-to-image, general editing, multi-character consistency, and mask editing. Independent Arena rankings confirm notable gains: 2.1 currently ranks 5th in text-to-image and 6th in single-image editing, the highest among Google’s models, though still trailing GPT Image 2.5 and GPT Image 2. Pricing has shifted overall: Standard API image output costs are $0.0336 (1K), $0.0504 (2K), and $0.0756 (4K), roughly half the price of Nano Banana 2. However, input pricing has risen from $0.50 per million tokens to $1.50, while text and thinking output pricing has increased from $3 to $7.50. The more reference images used and the higher the thinking intensity, the smaller the cost advantage from the lower image pricing.
Not financial or tax advice. PANews content is strictly educational and informational and is not investment advice, financial advice, tax advice, legal advice, or a solicitation to buy or sell any digital asset, security, or financial product. Do your own research and consult qualified advisers.
Disclosure. PANews may publish sponsored content, partner content, advertisements, affiliate links, event promotions, and market commentary involving Web3 projects, service providers, or financial products. PANews personnel, contributors, or affiliates may hold digital assets or other interests related to covered topics. See our Terms of Service.
A post on X says Google’s Nano Banana 2.1 is now available, offering high-quality image generation at lower prices across Google’s AI products. That would be a notable upgrade to one of the company’s most popular creative tools.
The paper trail does not back it up. As of October 6, 2026, Google has not confirmed any rollout of a model called Nano Banana 2.1. It is missing from AI Studio, the Gemini API, and the company’s pricing documentation.
What actually happened The name did appear somewhere official. Starting October 5, 2026, users spotted Nano Banana 2.1 in the model selector of Google Flow, the company’s AI video tool.
It did not stay long. The listing was pulled shortly after users began reporting it, and no announcement came with it.
Google Flow’s own model help page reinforces the point. It lists only the established Nano Banana variants and makes no mention of a 2.1 tier.
The claim about reduced prices also has nothing behind it in the documentation. Google has made no pricing changes, and nothing points to a new tier tied to the 2.1 reference.
The Nano Banana lineup that does exist Nano Banana is Google’s branding for its image models. They handle text-to-image generation, image editing, and conversational workflows where users refine a picture through back-and-forth prompts.
Three versions are currently in production. Each carries a friendlier consumer name and a more technical model identifier:
Nano Banana 2 runs as gemini-3.1-flash-image.
Nano Banana 2 Lite runs as gemini-3.1-flash-lite-image.
Nano Banana Pro runs as gemini-3-pro-image.
All three support output resolutions up to 4K, along with a range of aspect ratios. Pricing scales with resolution and usage.
At the bottom of the range, a 1K image from the Lite model costs approximately $0.0336. At the top, a 4K image from Pro runs $0.24.
The family has grown steadily this year. The two most recent additions arrived in February 2026 and June 2026, with Nano Banana 2 Lite becoming available on June 30, 2026.
Why the distinction matters For developers, the gap between a leaked name and a shipped model is practical, not academic. You cannot call a model through the Gemini API if it is not listed there. You also cannot budget around prices that have not been published.
Anyone building products on Nano Banana should keep planning around the three documented models and their published rates. Rewriting an integration or a cost forecast based on a vanished dropdown entry would be premature.
Until a listing appears in AI Studio, an entry in the Gemini API model list, and new rows in the pricing documentation, the lower prices described in the post on X remain a claim, not a rate card. Developers looking for cheaper image generation already have an option in Nano Banana 2 Lite, which is documented, priced, and available now.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.
Not financial or tax advice. PANews content is strictly educational and informational and is not investment advice, financial advice, tax advice, legal advice, or a solicitation to buy or sell any digital asset, security, or financial product. Do your own research and consult qualified advisers.
Disclosure. PANews may publish sponsored content, partner content, advertisements, affiliate links, event promotions, and market commentary involving Web3 projects, service providers, or financial products. PANews personnel, contributors, or affiliates may hold digital assets or other interests related to covered topics. See our Terms of Service.
Opinion: The practicality of liquidation heatmaps is vastly overestimated, and they should not play a major role in trading systems.
Benson Sun, founder of Coinkarma, says the practical value of liquidation heatmaps is vastly overestimated. Many traders now spot a large cluster of liquidations at a certain price level and immediately interpret it as "the price will definitely go to sweep this area". However, liquidation charts themselves rely heavily on assumptions. Most liquidation heatmaps work by taking historical market trading data and plugging in a set of hypothetical leverage multiples—such as 5x, 10x, 20x, 30x, and 50x—to reverse-engineer the price levels where those positions "might" be liquidated after trading. As a result, the dense liquidation zones users see do not actually represent an identical number of positions waiting to be liquidated there; too many variables are at play. Those traders may have already exited their positions. Some may add to their positions, some may reduce them, and others may top up margin. As long as position size, margin, or average cost changes, the liquidation heatmap shifts. The longer the time frame, the larger this error becomes. Therefore, the most useful window for liquidation heatmaps is roughly the last 24 hours to 3 days. Using liquidation clusters from 7 days, 30 days, or even months ago as current "magnet price levels" is meaningless. Liquidation heatmaps can serve as auxiliary information for short-term market structure analysis, but they should not play a major role in a trading system, as they easily create a false sense of certainty. There will always be liquidation clusters above and below the price. When prices rise, it is said they will first sweep the short positions above; when prices fall, it is said they will first sweep the long positions below; if prices fall first then rise, it is said they will first sweep longs then shorts. In any case, there is an after-the-fact explanation, and it is hard to falsify beforehand—this makes it not an analytical tool, but pure pseudoscience.
7 minutes ago
The yield on the 10-year U.S. Treasury note climbed to 5.234%, marking its highest level since mid-2007.
According to market data from BIT (bit.com), the U.S. 10-year Treasury yield has risen to 5.234%, hitting its highest level since mid-2007. The 30-year Treasury yield climbed to 5.542%, the highest since 2004. This indicates a significant rise in the market’s pricing of long-term risk, with growing concerns over persistent inflation, the Federal Reserve’s maintenance of high interest rates, as well as massive fiscal deficits and debt sustainability. Long-term borrowing costs for governments, corporations and consumers have been pushed up, which typically suppresses stock market valuations, tightens global financial conditions, and may curb economic growth to some extent. For more analysis, see the article "U.S. Long and Short-Term Treasury Yield Curves Nearing Inversion: Is a Recession Precursor Signal Activated?"
7 minutes ago
OKX's Flash Earn Lite adds Tether USDT; subscribe to ETH to share the 400,000 USDT reward pool.
Official announcement: OKX Flash Earn Lite will launch a Tether USDT "Stake-to-Earn" campaign from 15:00 UTC+8 on September 29, 2026 to 15:00 UTC+8 on October 4, 2026. During the event, users who subscribe ETH will be eligible to split a total of 400,000 USDT in airdrop rewards, with a minimum subscription threshold of 0.01 ETH. Pre-subscriptions will open at 15:00 UTC+8 on September 28. Users will earn simple interest during the pre-subscription period, while USDT airdrop rewards will be calculated from the official start of the event. Rewards are accrued hourly, and users can redeem their funds at any time; any ETH not actively redeemed after the event concludes will automatically continue to be subscribed to OKX's simple earning products.
7 minutes ago
Kimi K3.1 is reportedly in the backend deployment stage: Kimi-K3-1 has passed call tests, bringing it one step closer to its official release.
News flash from Dongcha Beating AI: AI model evaluation account Lumina said it discovered 'kimi-k3-1' in an official product model list from Kimi. Lumina subsequently tested each model individually; all other invalid models returned errors, while kimi-k3-1 was recognized by the backend and accepted requests.
7 minutes ago
Community members disclosed that the Compound Foundation misappropriated v2 DAI reserves.
Compound community member ugurmersin has alleged that the Compound Foundation misappropriated its v2 DAI reserves. According to ugurmersin, Proposal 536 submitted in February this year allocated approximately 8.42 million DAI to the Foundation for custody, with strict restrictions: the funds must be used solely for protocol operations, not for speculative trading, remain the property of the DAO, and cannot be used for the Foundation’s own expenditures. On-chain records show that the funds were transferred to a trading platform, exchanged for around 344,800 COMP, and delegated to the Foundation’s voting address roughly 58 minutes before the deadlines of Proposals 580 and 582. ugurmersin claims this move was intended to push nearly all DAO funds into TMC and pass a roughly $52 million V4 proposal favorable to the Foundation, while multiple parties publicly referred to the funds as "liquidity DAI" throughout the process. The Compound Foundation has not yet responded to the allegations.
7 minutes ago
Binance to List 5 New Stock Trading Assets Including BRUN
Binance’s stock trading will launch five new assets at 21:30 today, namely Boost Run (BRUN), Greenland Mines (GRML), Octave Intelligence (OCTV), StablecoinX (USDE), and Wise Group (WSE). Following the completion of trading and settlement (T+1 trading day), these stocks will qualify for Full Payment Securities Lending (FPSL).
US regulators have abruptly shut down a California bank in the sixth bank failure of 2026.
The Federal Deposit Insurance Corporation (FDIC) says the California Department of Financial Protection and Innovation closed Nano Banc of Irvine on Friday and appointed the FDIC as receiver.
The California DFPI says it seized the bank after a March 6th, 2026 capital order required tangible shareholders’ equity of at least 9.5% or an exit through sale, merger, or liquidation.
By September 22nd, tangible equity had fallen to about $5.6 million, or 0.82% of assets, against a 3% statutory floor, according to the state order.
The FDIC entered into a purchase and assumption agreement with Sunwest Bank of Sandy, Utah, which will assume substantially all deposits and acquire certain assets.
“Nano Banc’s sole branch will reopen as a branch of Sunwest Bank during its normal business hours on Monday, September 28, 2026. Depositors of Nano Banc will automatically become depositors of Sunwest Bank. The deposits assumed by Sunwest Bank will continue to be insured by the FDIC, so there is no need for customers to change their banking relationship.”
As of June 30th, 2026, Nano Banc reported about $736 million in assets and roughly $686 million in deposits.
Sunwest agreed to purchase about $476 million of the failed bank’s assets, with the FDIC retaining the rest for later sale.
Preliminary estimates show the failure will cost the Deposit Insurance Fund about $114 million, a figure expected to change as retained assets are sold.
The FDIC says customers keep immediate access to deposits, including by check, ATM, and debit card, and loan customers should keep making payments as usual.
An FDIC FAQ for the closure says no one lost money on deposits and that substantially all deposits, regardless of dollar amount, were transferred to Sunwest.
Holding company Nano Financial Holdings, Inc. was not included in the closing or receivership.
Earlier 2026 failures covered by The Daily Hodl include Metropolitan Capital Bank & Trust in Illinois, Community Bank and Trust – West Georgia, Kentland Federal Savings and Loan Association in Indiana, Small Business Bank in Kansas, and Tioga-Franklin Savings Bank in Pennsylvania. With Nano Banc, that makes six bank failures in 2026.
US-listed BNB treasury firm Nano Labs released a statement saying that the X account of its founder Jack Kong was hacked and used to publish false project information.
Relevant content
Goldman Sachs Asset Management: The Federal Reserve is unlikely to enter a sustained interest rate hike cycle.
Goldman Sachs Asset Management said in a report that it still does not expect the Federal Reserve to enter a sustained interest rate hike cycle. The firm noted: "We believe tariff and energy-related price pressures may subside, there are almost no signs of economic overheating, and inflation expectations remain anchored." The Federal Reserve's dot plot from last week's meeting signals one additional rate hike this year. According to LSEG data, money market pricing implies a cumulative 37 basis points of rate hikes across the Fed's remaining two meetings this year.
9 minutes ago
Bankless co-founder once again recommends buying ZEC and NEAR; the four tokens he previously held each surged more than 100%.
On-chain analyst Ai Yi (@ai_9684xtpa) wrote that Bankless co-founder David Hoffman is among the standout traders in the current market cycle. Hoffman liquidated his entire ETH position in May and reallocated funds to VVV, NEAR, ZEC, HYPE, and LIT in June. To date, four of the assets he disclosed purchasing have posted sharp gains, with LIT leading at a 293% increase. Specific entry costs and current returns are as follows: NEAR: ~$1.4 entry, up ~223%; ZEC: ~$560 entry, up ~173%; HYPE: ~$45 entry, up ~106%; LIT: ~$1.35 entry, up ~293%; VVV: entry cost undisclosed. Hoffman reiterated his bullish outlook on ZEC and NEAR earlier today. He stated, “ZEC in 2026 is what ETH was in 2021,” adding that if ZEC’s narrative can convince some Bitcoin holders, its price could continue to climb. He also labeled NEAR a “new consensus anchor for smart contracts,” noting its buying pressure is likely driven primarily by retail investors, though the token is also on an upward trajectory.
9 minutes ago
Meme coin GP on the Solana blockchain briefly saw its market capitalization exceed $30 million, surging over 101% in 24 hours.
According to GMGN data monitoring, the Solana-based meme coin RuneScape Gold (GP) briefly broke through $30 million in market capitalization, hitting an all-time high, and is now trading at $27.4 million, with a 24-hour increase of over 101%. GP is themed around the in-game gold of classic game *RuneScape*, paired with tokenized gold GLDx. Each GP transfer charges a 3% tax; accumulated fees are regularly converted by StonkFun and distributed proportionally to eligible holders in the form of GLDx, forming a mechanism of "earning gold price exposure rewards by holding tokens". It is a high-volatility meme asset that combines nostalgic game narratives, transaction tax dividends, and the concept of tokenized gold. BlockBeats reminds users that most meme coins have no practical use cases, experience significant price volatility, and require caution when investing.
9 minutes ago
A HYPE whale cut losses on $16.4 million worth of long position exposure, only retaining some spot positions for defense.
According to TradingBeats' monitoring, Hyperliquid address 0x4e23 liquidated all 178,800 HYPE long positions in batches this morning, worth approximately $16.45 million, with a loss of around $250,000. After the liquidation, the address only holds roughly 375 spot HYPE and 10,100 staked HYPE for defense, having temporarily removed its leveraged exposure. Currently, it still has buy orders worth about $6.38 million placed at the $87-$89 level, and sell orders totaling around $13.71 million at the $98-$101 level. Previously, the address went long on HYPE last night and exited all positions this morning; both of its recent two attempts to chase the rally ended in losses. The address also previously recovered most of its losses on SPCX US stock trades by repeatedly adjusting its position direction in a similar manner.
9 minutes ago
A prominent trader predicts Bitcoin could correct to $79,000, with the bull market set to continue after the adjustment concludes.
Well-known trader Doctor Profit stated that he has increased his Bitcoin short position to $13 million, expecting BTC to undergo a correction targeting $79,000. He will decide on his next moves based on price performance once Bitcoin hits that level, and emphasized that this short position does not reflect a bearish outlook on Bitcoin’s long-term trend. The trader believes Bitcoin’s bear market has ended, and the current pullback is merely a phase adjustment within the ongoing bull run; once the correction concludes, Bitcoin’s bull market will continue.
9 minutes ago
The Japanese yen has declined for two consecutive weeks, approaching the 160 threshold, and foreign exchange intervention risks are resurfacing.
With Japan’s holiday period drawing to a close, the risk of yen intervention is back in focus. The yen has fallen for two consecutive weeks, bringing it once again near the closely watched 160 level. Strategists believe the 160 level is once again a test of Japan’s tolerance for yen depreciation, as the currency has remained weak since the Bank of Japan’s (BOJ) September 18 policy meeting. While the BOJ has accelerated its tightening cycle, internal divisions exist within the committee, while the U.S. appears to be moving toward a more hawkish path. Carol Kong, currency strategist at the Commonwealth Bank of Australia (CBA), stated that if U.S. yields continue to rise and markets keep testing Japan’s resolve to defend the yen, USD/JPY could soon break through 160. A rapid break above this level would substantially raise the likelihood of official action, particularly given recent reports that Japan has conducted foreign exchange checks and the precedent of coordinated interventions. (Jinshi)
Ledger now lets Zcash (ZEC) holders keep private balances inside its own desktop app, Ledger Wallet. The update lands as ZEC trades 84% higher than a month ago.
Ledger makes hardware wallets, small devices that keep the keys to crypto funds offline. Until now, its users needed a separate third-party app to hold private ZEC.
Zcash (ZEC) Price Performance. Source: BeInCryptoWhat Changes for Zcash Holders on LedgerZcash offers two kinds of balance:
A transparent balance is public, like Bitcoin. A private, or shielded, balance hides the amounts and addresses involved. Ledger’s support page says one account can now hold both.
Private funds only appear if they sit in Ironwood, the new privacy pool Zcash launched in July. It replaced the old pool after researcher Taylor Hornby found a flaw there, as covered in the Ironwood upgrade.
Ledger Chief Technology Officer Charles Guillemet said the private data never leaves the user’s computer.
“Privacy here is not a server setting. To keep your shielded balance private, the scanning and the transaction building happen on your machine: your unified viewing key is stored locally and is never shared with anyone, including us,” Guillemet wrote.
A viewing key lets software read a wallet’s private history. Some simpler wallets send it to a server.
However, there are limits. Private ZEC cannot be swapped without first making it public, and the original Nano S cannot run the feature.
An older app from developer Zondax will be pulled on November 5, so its users must move their funds before then.
Where ZEC’s Rally StandsZEC traded at $1,516 as of this writing, down almost 3% in the last 24 hours. However, it is up 23% over seven days and over 84% in the last month, ranking ninth by market value.
Money has also flowed into Zcash funds. Zcash exchange-traded funds drew $98.2 million in the week to September 18, the largest weekly ETF inflow among 14 crypto products.
ZEC Spot ETF Flows. Source: SoSoValueSome backers see more room to run.
Zcash will replace bitcoin
Not because we want it to
but because it has to
We cannot live in a world where the mass adopted coin is also the public surveillance coin
This is how we avoid the Orwellian nightmare
— Thor Torrens (@ThorTorrens) September 22, 2026 ZEC hit a 24-hour high of $1,658.86 before sliding back below $1,530 at the time of writing. Ledger, meanwhile, left the choice to users, asking followers whether they hold ZEC shielded or transparent.
In brief ChatGPT Images 2.5 launched September 8 with sharper detail, more precise editing, and up to 50% lower latency than its predecessor Across six fresh test categories, Nano Banana 2 won three, with ChatGPT Images 2.5 taking the other ones The swing came down to a few checkable mistakes instead of very noticeable issues. OpenAI shipped ChatGPT Images 2.5 on September 8, and the pitch was specific: sharper details, richer textures, more natural lighting, and editing that actually respects what you didn't ask it to change.
The company says image-generation latency dropped by as much as 50% compared with Images 2.0, and two new models—GPT-Image-2.5 Flare and GPT-Image-2.5 Sunburst—now sit in the API, with Flare as the fast default and Sunburst built for premium editing precision.
Myriad: Which company will IPO next? Click to make your prediction.It's the direct sequel to a comparison we ran back in May, when GPT Image 2 and Nano Banana 2 traded wins across eight categories. GPT Image 2 took more of them but suffered from a distinctive oversharpening artifact whenever a prompt piled on too many constraints. That review closed with a specific question hanging over it: would OpenAI's next model fix that problem?
So we ran it back.
Same evaluator's eye, six categories carried over or adapted from earlier rounds, same prompts fed to both models. Google's side of the test uses Nano Banana 2—Google's name for Gemini 3.1 Flash Image—not the slower, more deliberate Nano Banana Pro.
This is a same-tier matchup: OpenAI's new fast-and-precise model against Google's fast-and-precise model.
What actually changed in GPT-Image 2.5OpenAI's image models have a pattern of shipping with a signature flaw. The model behind the original GPT Image 1 picked up a persistent warm, yellow color cast that the internet nicknamed the "piss filter"—a tint OpenAI never fully explained and never fully fixed.
GPT Image 2 traded that problem for a different one: feed it a prompt with too many stacked constraints and it would oversharpen the output into a crunchy, over-processed mess, thick with artifacts.
Neither flaw showed up in this round. Every image we generated with Images 2.5 held its color balance and its detail at full complexity, without the yellow cast or the crunchy over-sharpening that marked its two predecessors. That's a real, visible fix, and it shows most clearly in the steampunk and portrait tests—both are the most photographically coherent ChatGPT outputs we've tested across three model generations.
OpenAI also added product features that don't show up in a side-by-side image test but matter for workflow: Sketch, which lets you draw a rough layout directly in ChatGPT as a generation reference, prompt sharing, inline comments on specific image regions, and format templates for posters and merch. On the API side, quality tiers now run from low up through a new high and max, above where Images 2.0 topped out.
Here is how OpenAI’s new model stacks up against Google’s best offer.
Lettering density: the Kellerman's Hardware scene
This is the punishing one—a gritty 2 a.m. intersection where nearly every surface carries readable text: a ghost sign, spray-painted graffiti, vinyl storefront lettering, a torn concert poster, a stenciled curb, and a sticker-covered payphone.
Nano Banana 2 rendered almost everything cleanly. The one slip is a payphone sticker that duplicates its own text in a garbled way—a minor, easy-to-miss flaw.
Nano Banana 2: Kellerman's Hardware sceneChatGPT Images 2.5 added a detail Nano Banana skipped entirely: a lamppost with overlapping stapled flyers, torn and weathered, exactly as the prompt described.
ChatGPT Images 2.5: Kellerman's Hardware sceneBut the model's own street-art tag reads like "STILLL HERE" with an extra L, and the apostrophe in "KELLERMAN'S" on the ghost sign is missing or unreadable. For a model whose entire pitch is precision text rendering, two separate legibility slips in one image is a real miss.
Aesthetically speaking, OpenAI’s model generated a more realistic scene, but in terms of text it didn’t pay the same amount of care. Even if the image seems better, the text generation wasn’t as good as Google’s.
Winner: Nano Banana 2.
Spatial awareness: The steampunk clock tower
A demanding aerial composition test: a five-plane depth scene with a massive clocktower whose faces need to show different times in legible Roman numerals, plus six other text elements scattered from foreground to background.
ChatGPT Images 2.5 produced the more atmospheric image by a clear margin—visible steam rising off rooftops, a river cutting through the mid-ground, richer tonal range across the five depth planes. The letters are also easier to read in this generation.
ChatGPT Images 2.5: The steampunk clock towerNano Banana 2's atmosphere is flatter, but its two visible clock faces both render legible Roman numerals—XII, III, VI, IX—at similar hand positions which was not what the prompt asked for.
Nano Banana 2: The steampunk clock towerWinner: GPT Images 2.5, on the strength of following instructions better without sacrificing realism.
Illustration: The anime spirit medium
This prompt asks for a Studio Ufotable-style key visual: a girl mid-transformation into spiritual energy at a torii gate, with a nine-tailed kitsune fox and a Makoto Shinkai-painted twilight sky.
ChatGPT Images 2.5 delivers the best sky of any test we've run in this whole series—an actual sun disc, water reflection, and mountain silhouette that genuinely earns the Shinkai comparison. The asymmetric eyes land well too. Where it drifts from the prompt is the "dissolving into energy" instruction, which the model reinterprets as an electric-crackle effect running through her hair rather than the flowing, translucent dissolve described.
ChatGPT Images 2.5: The anime spirit mediumNano Banana 2's wispy blue-white energy trail is the closer literal match to that specific line. Neither model rendered a convincing nine-tailed fox.
Nano Banana 2: The anime spirit mediumWinner: ChatGPT Images 2.5, on overall visual impact, even with the interpretive drift.
Realism: The rooftop architect
A cinematic portrait with a long list of independent constraints: beige trench coat, round glasses, blueprints held in the left hand specifically, golden-hour lighting, shallow depth of field, film grain.
ChatGPT Images 2.5 produced gorgeous light—the sun disc is visible directly behind her, and the skin micro-texture is excellent. The skin is too smooth, but the overall scenery is very realistic, like taken with an analog camera.
ChatGPT Images 2.5: The rooftop architectInterestingly enough, adding some commands that would ordinarily make the photo bad quality actually end up increasing its realism. Here is a generation after adding keywords like “realistic, highlights, crushed shadows, uneven flash, blown out skin tones, candid moment, shot on a phone camera.”
ChatGPT Images 2.5: The rooftop architectNano Banana 2 keeps the fuller composition, with right-hand placement instead of left-handed, and adds a legible blueprint label—"PROJECT: 124 DUANE ST"—that most renders skip entirely.
Nano Banana 2: The rooftop architectWinner: Nano Banana 2, for one shot, GPT for repeated iterations.
Agentic research: The Bitcoin timeline
Both platforms can research a topic before rendering it, so we asked for a widescreen Bitcoin history timeline in kids-drawing style, with a strict bar on factual accuracy. This is the category that mattered most for the comparison, and it's where the biggest gap showed up.
ChatGPT Images 2.5 built a clean two-row infographic with specific dates throughout—and one of them is wrong. It labels 2023 as the year Bitcoin ETFs were approved in the U.S. The SEC actually approved the first spot Bitcoin ETFs on January 10, 2024—a full year later, even though the futures ETFs were actually approved that year, so it may be an interpretation issue.
ChatGPT Images 2.5: The Bitcoin timelineNano Banana 2's output is less structured but portrays similar events, which says something about how little variance the model shows on repeated prompts. That said, it hedges the ETF approval and the fourth halving into a "2023–2024" bracket instead of asserting a single wrong year, so nothing it states is technically false.
Nano Banana 2: The Bitcoin timelineWinner: Nano Banana 2. A confidently wrong date matters more in a category specifically testing whether "agentic reasoning" produces accurate output.
Abstract concepts: A prompt made of invented words
This one used a prompt built almost entirely from words that don't mean anything: "A woman eating shmfiyxl in Lyxin. Next to her, her Lymglsushing plays Lakishkark." There's no dictionary definition for the dish, the venue, the companion, or the game—each model has to invent one and then decide how to show its work.
ChatGPT Images 2.5 solved it by writing the nonsense straight into the scene as text. "Lyxin" glows on a neon sign above a sleek, futuristic restaurant. "Lakishkark" is printed on the box of a board game the woman's alien tablemate is playing. The made-up words stop being abstract the moment they become signage you can actually read.
ChatGPT Images 2.5: A prompt made of invented wordsNano Banana 2 took the opposite approach. It built a warm, culturally specific scene—a Guatemalan market stall, a woman in a traditional huipil, an orc-like creature playing a hybrid stringed-and-pipe instrument—and read "plays" as playing music rather than playing a game, a different but equally valid interpretation. Nothing in the frame ties back to "Lyxin" or "Lakishkark" specifically, though; the invented words never surface as visible text anywhere.
Nano Banana 2: A prompt made of invented wordsWinner: ChatGPT Images 2.5. Turning undefined concepts into legible labels is a more literal answer to a prompt that gives it nothing else to work with.
The verdict
Nano Banana 2 won three of six categories in this round, so it’s basically a tie. It will all depend on your expectations and the way you interact with the model. ChatGPT Images 2.5 also fixed the oversharpening problem that dogged its predecessor, and its illustration output this round is arguably the most visually striking single image either model has produced across two full rounds of testing.
What separates the two isn't overall quality but very tiny specific things: checkable misses, a spelling error in a lettering-heavy scene, a wrong year in a research-driven infographic, etc.
But in terms of overall aesthetics and quality they are pretty much on par.
Daily Debrief NewsletterStart every day with the top news stories right now, plus original features, a podcast, videos and more.
In brief ChatGPT Images 2.5 launched September 8 with sharper detail, more precise editing, and up to 50% lower latency than its predecessor Across six fresh test categories, Nano Banana 2 won three, with ChatGPT Images 2.5 taking the other ones The swing came down to a few checkable mistakes instead of very noticeable issues. OpenAI shipped ChatGPT Images 2.5 on September 8, and the pitch was specific: sharper details, richer textures, more natural lighting, and editing that actually respects what you didn't ask it to change.
The company says image-generation latency dropped by as much as 50% compared with Images 2.0, and two new models—GPT-Image-2.5 Flare and GPT-Image-2.5 Sunburst—now sit in the API, with Flare as the fast default and Sunburst built for premium editing precision.
Myriad: Which company will IPO next? Click to make your prediction.It's the direct sequel to a comparison we ran back in May, when GPT Image 2 and Nano Banana 2 traded wins across eight categories. GPT Image 2 took more of them but suffered from a distinctive oversharpening artifact whenever a prompt piled on too many constraints. That review closed with a specific question hanging over it: would OpenAI's next model fix that problem?
So we ran it back.
Same evaluator's eye, six categories carried over or adapted from earlier rounds, same prompts fed to both models. Google's side of the test uses Nano Banana 2—Google's name for Gemini 3.1 Flash Image—not the slower, more deliberate Nano Banana Pro.
This is a same-tier matchup: OpenAI's new fast-and-precise model against Google's fast-and-precise model.
What actually changed in GPT-Image 2.5OpenAI's image models have a pattern of shipping with a signature flaw. The model behind the original GPT Image 1 picked up a persistent warm, yellow color cast that the internet nicknamed the "piss filter"—a tint OpenAI never fully explained and never fully fixed.
GPT Image 2 traded that problem for a different one: feed it a prompt with too many stacked constraints and it would oversharpen the output into a crunchy, over-processed mess, thick with artifacts.
Neither flaw showed up in this round. Every image we generated with Images 2.5 held its color balance and its detail at full complexity, without the yellow cast or the crunchy over-sharpening that marked its two predecessors. That's a real, visible fix, and it shows most clearly in the steampunk and portrait tests—both are the most photographically coherent ChatGPT outputs we've tested across three model generations.
OpenAI also added product features that don't show up in a side-by-side image test but matter for workflow: Sketch, which lets you draw a rough layout directly in ChatGPT as a generation reference, prompt sharing, inline comments on specific image regions, and format templates for posters and merch. On the API side, quality tiers now run from low up through a new high and max, above where Images 2.0 topped out.
Here is how OpenAI’s new model stacks up against Google’s best offer.
Lettering density: the Kellerman's Hardware scene
This is the punishing one—a gritty 2 a.m. intersection where nearly every surface carries readable text: a ghost sign, spray-painted graffiti, vinyl storefront lettering, a torn concert poster, a stenciled curb, and a sticker-covered payphone.
Nano Banana 2 rendered almost everything cleanly. The one slip is a payphone sticker that duplicates its own text in a garbled way—a minor, easy-to-miss flaw.
Nano Banana 2: Kellerman's Hardware sceneChatGPT Images 2.5 added a detail Nano Banana skipped entirely: a lamppost with overlapping stapled flyers, torn and weathered, exactly as the prompt described.
ChatGPT Images 2.5: Kellerman's Hardware sceneBut the model's own street-art tag reads like "STILLL HERE" with an extra L, and the apostrophe in "KELLERMAN'S" on the ghost sign is missing or unreadable. For a model whose entire pitch is precision text rendering, two separate legibility slips in one image is a real miss.
Aesthetically speaking, OpenAI’s model generated a more realistic scene, but in terms of text it didn’t pay the same amount of care. Even if the image seems better, the text generation wasn’t as good as Google’s.
Winner: Nano Banana 2.
Spatial awareness: The steampunk clock tower
A demanding aerial composition test: a five-plane depth scene with a massive clocktower whose faces need to show different times in legible Roman numerals, plus six other text elements scattered from foreground to background.
ChatGPT Images 2.5 produced the more atmospheric image by a clear margin—visible steam rising off rooftops, a river cutting through the mid-ground, richer tonal range across the five depth planes. The letters are also easier to read in this generation.
ChatGPT Images 2.5: The steampunk clock towerNano Banana 2's atmosphere is flatter, but its two visible clock faces both render legible Roman numerals—XII, III, VI, IX—at similar hand positions which was not what the prompt asked for.
Nano Banana 2: The steampunk clock towerWinner: GPT Images 2.5, on the strength of following instructions better without sacrificing realism.
Illustration: The anime spirit medium
This prompt asks for a Studio Ufotable-style key visual: a girl mid-transformation into spiritual energy at a torii gate, with a nine-tailed kitsune fox and a Makoto Shinkai-painted twilight sky.
ChatGPT Images 2.5 delivers the best sky of any test we've run in this whole series—an actual sun disc, water reflection, and mountain silhouette that genuinely earns the Shinkai comparison. The asymmetric eyes land well too. Where it drifts from the prompt is the "dissolving into energy" instruction, which the model reinterprets as an electric-crackle effect running through her hair rather than the flowing, translucent dissolve described.
ChatGPT Images 2.5: The anime spirit mediumNano Banana 2's wispy blue-white energy trail is the closer literal match to that specific line. Neither model rendered a convincing nine-tailed fox.
Nano Banana 2: The anime spirit mediumWinner: ChatGPT Images 2.5, on overall visual impact, even with the interpretive drift.
Realism: The rooftop architect
A cinematic portrait with a long list of independent constraints: beige trench coat, round glasses, blueprints held in the left hand specifically, golden-hour lighting, shallow depth of field, film grain.
ChatGPT Images 2.5 produced gorgeous light—the sun disc is visible directly behind her, and the skin micro-texture is excellent. The skin is too smooth, but the overall scenery is very realistic, like taken with an analog camera.
ChatGPT Images 2.5: The rooftop architectInterestingly enough, adding some commands that would ordinarily make the photo bad quality actually end up increasing its realism. Here is a generation after adding keywords like “realistic, highlights, crushed shadows, uneven flash, blown out skin tones, candid moment, shot on a phone camera.”
ChatGPT Images 2.5: The rooftop architectNano Banana 2 keeps the fuller composition, with right-hand placement instead of left-handed, and adds a legible blueprint label—"PROJECT: 124 DUANE ST"—that most renders skip entirely.
Nano Banana 2: The rooftop architectWinner: Nano Banana 2, for one shot, GPT for repeated iterations.
Agentic research: The Bitcoin timeline
Both platforms can research a topic before rendering it, so we asked for a widescreen Bitcoin history timeline in kids-drawing style, with a strict bar on factual accuracy. This is the category that mattered most for the comparison, and it's where the biggest gap showed up.
ChatGPT Images 2.5 built a clean two-row infographic with specific dates throughout—and one of them is wrong. It labels 2023 as the year Bitcoin ETFs were approved in the U.S. The SEC actually approved the first spot Bitcoin ETFs on January 10, 2024—a full year later, even though the futures ETFs were actually approved that year, so it may be an interpretation issue.
ChatGPT Images 2.5: The Bitcoin timelineNano Banana 2's output is less structured but portrays similar events, which says something about how little variance the model shows on repeated prompts. That said, it hedges the ETF approval and the fourth halving into a "2023–2024" bracket instead of asserting a single wrong year, so nothing it states is technically false.
Nano Banana 2: The Bitcoin timelineWinner: Nano Banana 2. A confidently wrong date matters more in a category specifically testing whether "agentic reasoning" produces accurate output.
Abstract concepts: A prompt made of invented words
This one used a prompt built almost entirely from words that don't mean anything: "A woman eating shmfiyxl in Lyxin. Next to her, her Lymglsushing plays Lakishkark." There's no dictionary definition for the dish, the venue, the companion, or the game—each model has to invent one and then decide how to show its work.
ChatGPT Images 2.5 solved it by writing the nonsense straight into the scene as text. "Lyxin" glows on a neon sign above a sleek, futuristic restaurant. "Lakishkark" is printed on the box of a board game the woman's alien tablemate is playing. The made-up words stop being abstract the moment they become signage you can actually read.
ChatGPT Images 2.5: A prompt made of invented wordsNano Banana 2 took the opposite approach. It built a warm, culturally specific scene—a Guatemalan market stall, a woman in a traditional huipil, an orc-like creature playing a hybrid stringed-and-pipe instrument—and read "plays" as playing music rather than playing a game, a different but equally valid interpretation. Nothing in the frame ties back to "Lyxin" or "Lakishkark" specifically, though; the invented words never surface as visible text anywhere.
Nano Banana 2: A prompt made of invented wordsWinner: ChatGPT Images 2.5. Turning undefined concepts into legible labels is a more literal answer to a prompt that gives it nothing else to work with.
The verdict
Nano Banana 2 won three of six categories in this round, so it’s basically a tie. It will all depend on your expectations and the way you interact with the model. ChatGPT Images 2.5 also fixed the oversharpening problem that dogged its predecessor, and its illustration output this round is arguably the most visually striking single image either model has produced across two full rounds of testing.
What separates the two isn't overall quality but very tiny specific things: checkable misses, a spelling error in a lettering-heavy scene, a wrong year in a research-driven infographic, etc.
But in terms of overall aesthetics and quality they are pretty much on par.
Daily Debrief NewsletterStart every day with the top news stories right now, plus original features, a podcast, videos and more.
OpenAI dropped its newest image generation model on September 8, and the early results suggest it’s a genuine leap forward. ChatGPT Images 2.5 cuts generation latency by up to 50% compared to its predecessor while producing noticeably sharper visuals, richer textures, and more natural lighting. The platform now powers over 3 billion image generations per week.
How the two models stack up Google’s competing model, Nano Banana 2, built on Gemini 3.1 Flash Image, arrived earlier this year on February 26. It was designed around fast generation and high-fidelity outputs, and it does those things well. In structured tasks requiring tight prompt precision, Google’s offering still holds an edge.
But when the prompts get more creative, OpenAI pulls ahead. Independent testing across six categories found that ChatGPT Images 2.5 outperforms Nano Banana 2 in tasks emphasizing natural blending, visual depth, and editing reliability. The difference is most apparent in practical scenarios: multi-turn editing sessions where users refine an image through several rounds of conversation.
One standout feature is in-chat sketching, where users can draw rough outlines and have the AI interpret them into polished visuals. ChatGPT Images 2.5 handles this with considerably more nuance, maintaining subject integrity across edits rather than drifting into uncanny territory after a few rounds of refinement.
Two API flavors, one clear strategy OpenAI didn’t release just one API variant. It launched two: GPT-Image-2.5 Flare, which balances speed and quality, and GPT-Image-2.5 Sunburst, optimized for maximum precision. Both models are available across ChatGPT, ChatGPT Work, and Codex.
On the safety and provenance front, both OpenAI and Google now embed C2PA metadata in generated images. Both models also support reference images and multiple aspect ratios.
The competitive landscape is getting interesting For businesses building on top of these APIs, the performance differential matters. A 50% reduction in latency isn’t just a nice benchmark number. It translates directly into lower compute costs, faster user experiences, and the ability to handle higher volumes without proportional infrastructure spending.
Google’s advantage in structured prompt adherence keeps it competitive for enterprise use cases where consistency and specification compliance matter more than artistic quality. The multi-turn editing improvements are particularly significant because they reduce the number of generation attempts needed to reach a final output.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.
Bonk Guy: PONS and USELESS will both surpass a $1 billion market cap this cycle.
Well-known trader Bonk Guy stated in his prediction that PON will become the first utility token to reach a $1 billion market cap in this cycle, while USELESS will be the first meme coin to hit that milestone in the same period. GMGN data shows PON currently has a market cap of $650 million, having previously peaked at $990 million. USELESS reached a cycle high of $330 million, and now stands at a $250 million market cap. BlockBeats reminds users that token prices are highly volatile, so investment should be approached with caution.
6 minutes ago
INDEX and COOPERATIVE have been added to Robinhood’s tradable assets.
INDEX and COOPERATIVE have been added to Robinhood’s tradable assets, allowing users to trade them directly via the Robinhood App. INDEX (The Index) is an RWA protocol token built on Robinhood Chain. After users hold a certain amount of INDEX, the protocol uses 3% of generated transaction fees to purchase a basket of tokenized U.S. stocks (including AAPL, NVDA, TSLA, etc.), which are automatically airdropped to holders’ wallets every 15 minutes—no staking or manual claiming required.
6 minutes ago
Robinhood Chain ecosystem token INDEX sees short-term rally, market cap surges past $45 million.
According to GMGN data, the Robinhood Chain ecosystem token INDEX has seen a short-term rally, pushing its market capitalization above $45 million, with a 24-hour gain of 17.8% and trading volume of $5.5 million over the same period. INDEX has been added to Robinhood’s tradable assets, allowing users to trade directly via the Robinhood App. INDEX (The Index) is an RWA protocol token on Robinhood Chain. After holding a certain amount, the protocol uses 3% of transaction fees to purchase a basket of tokenized U.S. equities (such as AAPL, NVDA, TSLA, etc.), which are automatically airdropped to holders’ wallets every 15 minutes—no staking or manual claiming is required. BlockBeats reminds users that such tokens are highly volatile, so investment should be approached with caution.
6 minutes ago
Grayscale: Zcash mining profitability is significantly higher than Bitcoin’s, with mining activity surging more than 2.5 times this year.
Grayscale Research Director Zach Pandl published an analysis noting that Zcash (ZEC) is a cryptocurrency similar to Bitcoin but with additional privacy features, with both networks secured via proof-of-work (PoW) mining. Driven by Zcash’s strong token price performance, its mining has become highly profitable. While Bitcoin’s network is far larger—with daily total miner rewards of roughly $35 million versus Zcash’s ~$2 million—Zcash is more favorable for individual miners. Grayscale estimates that Zcash offers around twice the daily reward per mining rig and four times the reward per megawatt-hour (MWh) compared to Bitcoin, though their mining hardware is not interchangeable. Additional data shows Zcash’s current per-MWh mining yield is even higher than that of some AI/high-performance computing cloud services. This higher profitability is spurring more mining activity: measured by standard hashrate metrics, Zcash’s total network mining activity has grown over 2.5 times this year. This forms a virtuous cycle: higher token prices drive more mining, which enhances network security, and stronger security in turn helps sustain elevated prices. Grayscale concludes that at current valuation levels, Zcash mining is very profitable, attracting more hashrate and supporting network security.
6 minutes ago
STORJ surged 60.3% in the past 24 hours, and both Binance and Upbit have delisted it.
According to HTX market data, STORJ has surged 60.3% in the past 24 hours, currently trading at $0.04588. Binance delisted STORJ on September 3, while South Korea’s largest crypto exchange Upbit will delist STORJ on September 14. On July 27, Storj Labs, parent company of decentralized cloud storage project Storj, announced it had filed for voluntary Chapter 11 bankruptcy protection under U.S. bankruptcy code to resolve historical debts and sustain operations, with plans for the restructured firm to be jointly owned by management, the community, STORJ token holders and investors. Founded in 2014, Storj is a decentralized storage network where users provide cloud storage services to global clients by sharing unused hard drive space, using STORJ tokens for incentives and payments.
6 minutes ago
Crypto-related stocks extended their gains in early US stock trading, with BitMine rising 9.01%.
According to market data from BIT (bit.com), crypto-related stocks extended their gains at the start of U.S. trading: Strategy (MSTR) rose 5.89%, Coinbase (COIN) gained 5.29%, Circle (CRCL) increased 5.18%, BitMine Immersion (BMNR) climbed 9.01%, and SharpLink Gaming (SBET) jumped 10.02%.
Not financial or tax advice. PANews content is strictly educational and informational and is not investment advice, financial advice, tax advice, legal advice, or a solicitation to buy or sell any digital asset, security, or financial product. Do your own research and consult qualified advisers.
Disclosure. PANews may publish sponsored content, partner content, advertisements, affiliate links, event promotions, and market commentary involving Web3 projects, service providers, or financial products. PANews personnel, contributors, or affiliates may hold digital assets or other interests related to covered topics. See our Terms of Service.
Beating AI Express: Tencent Hunyuan has open-sourced AuK, a 1.5-billion-parameter speech generation and editing model, which the official dubs the "audio version of Nano Banana". The model integrates functions including voice cloning, word editing, emotion switching, accent removal, speed and pitch adjustment, noise reduction, and multi-person/music separation, all controllable via natural language. AuK can directly modify existing recordings: it only adjusts incorrect words without re-recording the entire segment, alters emotion/timbre/accent while keeping content unchanged, edits lyrics, removes laughter/coughs, and converts normal speech to whispers. Notably, its voice cloning does not require a reference transcript—only a sample audio and new text are needed. In official tests, AuK leads in multiple speech generation and general speech editing benchmarks, scoring 49.73 on SpeechEditBench, while second-place Ming-UniAudio scores 28.70. The fast variant AuK-Flash, after distillation, needs just 4 inference steps and runs ~4.5x faster. However, the paper notes AuK cannot yet reliably understand all arbitrary natural language inputs, still relying on a Prompt Enhancer to identify tasks, organize parameters and rewrite instructions. Code and model weights are now public under the MIT license.
Relevant content
Oracle's Q1 results beat expectations, climbing more than 5% in after-hours trading.
Oracle has announced its fiscal 2027 first quarter results: adjusted earnings per share (EPS) stood at $1.92, beating the market consensus estimate of $1.75; adjusted total revenue hit $19.35 billion, also exceeding the $19.13 billion forecast. By business segment, cloud infrastructure (IaaS) revenue reached $7.39 billion, surpassing expectations of $7.19 billion; software support revenue totaled $4.9 billion, slightly above the $4.88 billion estimate; software license revenue came in at $655 million, however, missing the $718.8 million forecast. Boosted by the strong results and cloud business performance, Oracle (ORCL) saw its U.S. stock rise more than 5% in after-hours trading. In recent years, Oracle has been accelerating its heavy bet on AI cloud infrastructure. At the end of its last fiscal year, the company’s remaining performance obligations (RPO) reached $638 billion, up 363% year-over-year, with large-scale AI cloud contracts acting as a key driver of order growth.
1 seconds ago
European regulators warn Polymarket and Kalshi that they may lack the necessary licenses to serve EU users.
The European Securities and Markets Authority (ESMA) released its latest risk report, stating that major prediction market platforms including Polymarket and Kalshi currently lack the required authorization to offer event contracts to EU users under normal circumstances. ESMA noted that the regulatory nature of event contracts depends on the specific product, which may be classified as financial instruments, crypto assets subject to the MiCA framework, or fall under gambling regulations of EU member states. If categorized as financial instruments, they typically qualify as derivatives and may be subject to EU retail restrictions on binary options. ESMA also questioned the current EU region restrictions imposed by Polymarket and Kalshi, pointing out that the two platforms only restrict access to some member states, and it remains unclear why all EU countries have not been included in the restricted list, while there is also a risk of VPNs being used to bypass geographic restrictions. In July this year, France ordered internet service providers to block Polymarket. In addition, ESMA warned that the AI boom could become a new risk transmission channel for the crypto market. If AI investment returns fall short of expectations or there is a sell-off in tech stocks, large investors may sell highly liquid risky assets including crypto assets to raise cash. ESMA data shows that Bitcoin fell 35% in the first half of 2026, with outflows from US spot Bitcoin ETFs exceeding $5.5 billion over the same period. ESMA has previously clarified that if event contracts are classified as financial instruments, their distribution in the EU requires an investment firm license and may be subject to retail bans on binary options.
1 seconds ago
The US SEC intends to make blockchain the official record for securities ownership, marking a key regulatory breakthrough for tokenized stocks.
The U.S. Securities and Exchange Commission (SEC) recently proposed reforms to rules for registered transfer agents, aiming to explicitly allow electronic databases—including blockchain—as official records of securities ownership. Currently, tokenized securities often have both on-chain token ledgers and off-chain official shareholder registers; the need to continuously reconcile these two systems increases operational costs and legal risks. If the new rules are approved, blockchain is expected to serve as the primary record for securities, eliminating the need for tokenized securities to maintain duplicate off-chain ownership records. SEC Chair Paul Atkins stated that the reforms aim to align transfer agent rules with advancements in electronic record-keeping and blockchain technology. However, this does not mean tokenized securities will become fully permissionless assets. Issuers and transfer agents must still comply with regulatory requirements such as investor identity verification, transfer restrictions, succession protocols, and shareholder notifications; blockchain will primarily function as a legally recognized ownership record infrastructure. The SEC previously noted that, provided they meet federal securities laws’ requirements for record-keeping, reporting, and audits, registered transfer agents may use distributed ledgers as official shareholder registers without maintaining additional off-chain "digital twin" records. The rule proposal is open for public comment for 60 days. If finalized, blockchain will further evolve from a "technical shell" for tokenized securities to a core component of U.S. securities market infrastructure, removing key legal and operational barriers to the full on-chain migration of traditional assets like stocks and funds.
1 seconds ago
Trump to distribute $500 healthcare "refunds" to nearly one million people.
On September 10, the Trump administration announced it would issue a one-time $500 "refund" to nearly 1 million enrollees in the Affordable Care Act (Obamacare), with disbursements scheduled for October—just weeks ahead of the U.S. midterm elections. Eligible consumers are primarily those who purchased Obamacare plans via HealthCare.gov, have incomes roughly four times the federal poverty level, and did not receive federal subsidies, covering 30 states. The White House said the funds are refunds for unused health insurance marketplace operating fees previously collected by the federal government, arguing these costs were ultimately passed on to consumers. However, the policy has sparked controversy. Health policy experts point out that the Biden administration actually lowered ACA marketplace user rates, and such fees are not the main driver of this year’s sharp premium hikes. Meanwhile, after additional health insurance subsidies expired, many consumers face thousands of dollars in premium increases this year, with the $500 refund covering only a small portion of that cost. Notably, Trump’s refund announcement comes as midterm elections near. Earlier, he also proposed that if the Republican Party holds onto control of Congress in the November midterms, it would distribute $5,000 "dividends" to U.S. households. As healthcare, cost of living and other issues remain top voter concerns, the Trump administration is ramping up policies that directly deliver economic benefits to the public.
1 seconds ago
Codex Head of Product: Suspends new subscriptions for the $200 Pro plan.
Beating AI Express News: OpenAI Codex product lead Tibo stated, "To ensure existing users get an excellent experience and can continue using Astra, we are pausing new subscriptions to the $200 monthly Pro plan. This plan puts the most strain on our systems, so we want to make the smallest adjustment possible to keep access open to as many users as we can. All other plans and API services remain fully operational. Existing accounts will not be affected at all, and we are working to add more capacity as quickly as possible."
1 seconds ago
The three major U.S. stock indexes closed lower, with Apple rising more than 3% and PURR falling 8.49%.
US stocks closed Thursday: The Dow Jones fell 0.6%, the S&P 500 dropped 0.58%, and the Nasdaq declined 0.65%. Nvidia (NVDA.O) fell 2.2%, Intel (INTC.O) dropped 5.5%, SK Hynix (SKHY.O) declined 5.2%, while Apple (AAPL.O) gained 3.5%. Among crypto-related concept stocks: MSTR fell 3.12%, CRCL dropped 2.82%, COIN declined 1.40%, SBET fell 1.29%, PURR plunged 8.49%, and HOOD fell 1.69%.
Beating AI Express News: Leaker Lyra claims Google’s next Pro model already has its first checkpoint (a saved model state version from training) and is set to be unveiled in October. Lyra also predicts a new Flash-Lite launch in September, alongside an update to Nano Banana 2 Lite. The final name for the new Pro model remains unconfirmed; some community members speculate it could be Gemini 4 Pro, while others believe it may be another Pro variant.
CZ: Some hot money is flowing back from the AI sector to the crypto market, and the crypto industry will not disappear.
Binance founder CZ published a post noting that some "hot money" is flowing back from the AI sector to the crypto market. Currency-related industries will not disappear, as both individuals and AI will still need currency in the future.
1 hours ago
Ansem: Crypto Market Remains in the Early Stage of a Bull Run, Retail Investors Are Entering with More Capital
Crypto KOL Ansem posted an article stating that the crypto market is still in the early stages of a bull run, and the key to generating returns at this stage is to identify assets with asymmetric upside while tolerating short-term volatility. Over the past two years, rotating between meme coins and new trading pairs has been the dominant strategy, with lower valuation caps leading traders to favor short-term holdings; however, in a bull market, high-quality assets offer greater upside potential, so extending holding periods after careful selection may prove more advantageous. Ansem believes retail investors are entering the crypto market with more capital. The growth of mobile users on Pump.fun and Fomo, as well as Robinhood Chain’s ongoing efforts to convert stock traders to on-chain activities, all indicate that market liquidity may increase in the future. New users pay relatively less attention to market capitalization changes, so tokens that gain widespread traction may receive stronger capital inflows. He also noted that the trend toward short-form video has led fewer and fewer investors to read project whitepapers or research token differences, which in turn creates opportunities for those willing to build a complete investment thesis and exercise patience. However, traders still need to set criteria for when they are wrong, review the reasons for missing out on high-growth assets, and define conditions for re-entering the market after selling too early.
1 hours ago
Arbitrum DAO generated $6.19 million in revenue in the first half of the year, with Robinhood Chain emerging as a new revenue source.
An unaudited report released by the Arbitrum Foundation shows that Arbitrum DAO generated $6.19 million in revenue in the first half of 2026, with sources including Arbitrum One transaction fees, Timeboost sequencing priority auctions, scaling program licensing fees, and treasury management returns. The protocol’s gross profit margin exceeded 97%, and non-ARB treasury assets stood at $125 million as of the end of June. In H1 2026, Arbitrum processed a total of 478 million transactions, accounting for roughly 18% of its cumulative lifetime total of 2.7 billion transactions; monthly average stablecoin transfer volume surpassed $70 billion, and the number of stablecoin holders rose 40% to 10.5 million. Additionally, Arbitrum has deployed over 2,000 tokenized RWAs. Robinhood Chain, built on Arbitrum’s tech stack, launched its mainnet on July 1, contributing $360,000 in licensing fees to the DAO that month, making up 35% of its monthly revenue. On September 1, Robinhood Chain hit daily fees of $3.75 million, decentralized exchange (DEX) volume exceeding $1.5 billion, and total value locked (TVL) of over $750 million.
1 hours ago
Agent of "BTC OG Insider Whale": Bitcoin has held the $76,600 support level; if it breaks through $79,000, it could test higher highs.
Garrett Jin, the representative of the "BTC OG Insider Whale", stated in a post that Bitcoin (BTC) has held the critical level of $76,600. If BTC climbs further above $79,000, the price may attempt to form a higher high. However, even if BTC does post a higher high, this would still not be sufficient to confirm a genuine breakout in the market.
1 hours ago
NVIDIA rises nearly 5%, with its current market capitalization standing at $5.49 trillion.
According to market data from BIT (bit.com), NVIDIA's stock rose 4.82%, with its current market capitalization standing at $5.49 trillion.
1 hours ago
Making money from FOMO? The peak APR for the JINQIAN liquidity pool (LP) on Uniswap hit 83,832%.
Tonight, the "short squeeze on underlying stocks" narrative surrounding JINQIAN/FAMI on Robinhood’s blockchain has been denied by the parties involved, putting an end to the on-chain hype. However, a review of the entire incident shows that the biggest gains were not only made by top on-chain traders; smart money also reaped substantial profits by setting up JINQIAN trading pair liquidity provider (LP) pools on Uniswap. According to data from Uniswap’s official page, during JINQIAN’s first rally from a $7 million market cap to $60 million, the peak APR of its main trading pair JINQIAN/ETH hit 83,832%, and has since dropped to 79,708%. As of press time, the JINQIAN/USDG trading pair still has a 6% transaction fee, with the pool’s annualized APR reaching as high as 126,440%. This means that if traders bought JINQIAN tokens during the rally, then established liquidity pools at higher price levels, timely collected LP fees, and sold off part of their JINQIAN holdings, their risk-reward ratio would be far higher than that of simply holding JINQIAN tokens. However, current market sentiment has become overly FOMO, and on-chain scams are on the rise. Even setting up LP pools for individual popular meme coins cannot fully avoid risks from token price fluctuations, so users should exercise caution with their investments.
Not financial or tax advice. PANews content is strictly educational and informational and is not investment advice, financial advice, tax advice, legal advice, or a solicitation to buy or sell any digital asset, security, or financial product. Do your own research and consult qualified advisers.
Disclosure. PANews may publish sponsored content, partner content, advertisements, affiliate links, event promotions, and market commentary involving Web3 projects, service providers, or financial products. PANews personnel, contributors, or affiliates may hold digital assets or other interests related to covered topics. See our Terms of Service.
Nvidia just made its cheapest path into robotics AI significantly more powerful. The company has released the Jetson Orin Nano Super Developer Kit, an upgraded version of its entry-level edge computing platform that pushes AI performance from 40 TOPS to 67 TOPS, all while keeping the price at $249.
That 1.7x performance jump is the kind of upgrade that usually comes with a new product generation and a higher price tag. Instead, Nvidia managed to squeeze it out of the same hardware platform, meaning existing Orin Nano owners can unlock the boost through a JetPack software update. No new silicon required.
What the Jetson Orin Nano Super actually does The original Jetson Orin Nano Developer Kit debuted at GTC 2023 with up to 40 TOPS of performance. TOPS stands for Tera Operations Per Second, which measures how many trillions of basic math operations a chip can crunch every second. At 67 TOPS, the Super variant enters territory where it can actually handle the transformer-based AI models that have reshaped the industry over the past two years.
The Jetson Orin Nano Super is specifically optimized for transformer architectures, the same family of models behind large language models, vision-language models, and vision transformers. In practical terms, that means a small robotics platform can now run sophisticated AI for visual recognition, natural language understanding, and multimodal reasoning without needing a connection to cloud infrastructure.
Nvidia is positioning the kit for use across manufacturing, logistics, and healthcare. The energy efficiency of the platform makes it viable for mobile robots and embedded systems where power consumption matters as much as raw performance.
A $249 entry point changes the math Nvidia has described the Jetson Orin Nano Super as the most affordable generative AI supercomputer for edge devices. Robotics engineers at startups can prototype AI-powered devices without a massive hardware budget. University labs can equip students with capable development platforms at a fraction of what professional-grade AI hardware typically costs.
The JetPack software upgrade path is arguably just as important as the price. Developers who already invested in the original Orin Nano platform don’t have to buy new hardware to access Super-level performance.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.
Aptiv PLC just deepened its relationship with Nvidia, and the stock market noticed. The automotive and industrial technology company announced an expanded collaboration to deliver production-ready edge AI platforms built on Nvidia’s Jetson ecosystem, and shares responded with a 6.2% intraday pop.
What Aptiv and Nvidia are actually building The collaboration targets edge AI platforms, systems that process data locally rather than shipping everything to the cloud. At the hardware level, the partnership aligns with Nvidia’s Jetson Orin Nano 2 robotics computer. This updated variant delivers up to twice the inference performance of its predecessor while cutting power consumption by roughly 40% at comparable loads. The Orin Nano series can push up to 67 TOPS (tera operations per second). That’s enough horsepower for real-time object detection, path planning, and sensor fusion, the core building blocks of physical AI in robotics and automotive applications.
Aptiv’s contribution sits on the software and compliance side. The company is providing Cyber Resilience Act-ready Yocto platforms and enterprise Linux support, which translates to pre-configured operating systems that meet European cybersecurity regulations out of the box.
Why the production gap matters Aptiv offers comprehensive software support, security features, lifecycle management, and commercial deployment tools designed to reduce fragmentation across deployments. In industries like automotive, where a single platform might need to function reliably for 15 years, managing software updates and security vulnerabilities becomes a significant operational challenge.
The sectors in play here are broad: robotics, industrial automation, and automotive.
Market context The 6.2% share price jump followed the June 1 announcement. Some market observers have speculated about broader momentum that could push gains as high as 25%, though that projection depends on commercial uptake. Notably, Aptiv did not explicitly endorse the new Orin Nano 2 variant in its own communications, though the alignment of its production support with Nvidia’s advancing hardware signals collaborative intent. Design wins in automotive and industrial sectors typically take 12 to 24 months to translate into meaningful revenue.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.
Nvidia shares rise 1.88% as Jetson Orin Nano 2 boosts edge computing power. Jetson Orin Nano 2 doubles inference performance while reducing overall power use. Matic, Wing and other partners begin adopting Nvidia’s new robotics platform. Jetson Orin Nano 2 delivers 78 TOPS, 8GB memory and a powerful eight-core Arm CPU. Nvidia strengthens its robotics ecosystem as over three million developers build. Nvidia (NVDA) shares gained 1.88% to $212.40 after the company introduced its new Jetson Orin Nano 2 robotics computer. The stock recovered from an earlier pullback after briefly trading above $214 during the session. Meanwhile, Nvidia positioned the new system as a stronger entry-level platform for robotics and edge computing.
NVIDIA Corporation, NVDA
Nvidia Expands Edge Computing With Jetson Orin Nano 2 Nvidia designed Jetson Orin Nano 2 for developers building robots, drones, vision systems, and other autonomous machines. The platform brings stronger computing performance into a compact and power-efficient design for edge applications. Consequently, developers can process language, images, and environmental data directly on connected devices.
The system delivers 78 trillion operations per second and includes 8GB of memory with an eight-core Arm processor. Nvidia says the platform provides twice the inference performance of the previous Jetson Orin Nano Super. Higher memory bandwidth and improved Tensor Cores support the performance increase without changing the compact hardware format.
Meanwhile, Jetson Orin Nano 2 also improves energy efficiency for systems that operate under strict power limits. The computer uses 40% less power when matching its predecessor’s performance in 15-watt operating mode. Therefore, Nvidia can target smaller robots and drones that require stronger computing without significantly increasing energy use.
Robotics Companies Begin Testing Nvidia’s New Platform Several technology companies have started adopting or evaluating Jetson Orin Nano 2 for commercial edge computing applications. Cognex, Doosan Bobcat, Matic, and Wing rank among the companies exploring the new hardware. Their projects cover industrial systems, household robotics, drone delivery, and advanced machine vision.
Matic plans to use the platform to improve its home cleaning robots with stronger real-time processing capabilities. The company wants its robots to handle mapping, gesture detection, navigation, and conversational functions more efficiently. Additionally, local processing can help robots understand changing home environments without relying heavily on remote computing infrastructure.
Wing also plans to evaluate the new system for its delivery drones and autonomous operations. The Alphabet subsidiary already uses Jetson Orin Nano Super and Nvidia software across its drone fleet. Jetson Orin Nano 2 could support faster perception and reasoning while reducing power requirements during delivery operations.
Nvidia Strengthens Its Robotics Developer Ecosystem Nvidia has built a large robotics ecosystem around Jetson hardware and its supporting software tools. More than three million developers currently use the company’s robotics technology for physical computing applications. The latest platform gives this community another lower-cost option for developing smaller autonomous systems.
Jetson Orin Nano 2 supports large language and vision models designed for memory-efficient processing at the edge. Developers can use Nvidia’s software stack alongside models such as Cosmos, Nemotron, Gemma 4, and Qwen 3. This approach allows machines to process advanced models locally while reducing dependence on cloud infrastructure.
Hardware partners are also developing carrier boards, software, systems, and reference designs around the new platform. Companies supporting the launch include AAEON, ADLINK, Advantech, Aetina, Aptiv, Connect Tech, and Seeed Studio. Meanwhile, Nvidia continues expanding beyond data centers as robotics becomes another important market for its computing technology.
NANO Nuclear Energy and Tillman Global Holdings have entered a strategic commercial framework that positions NANO’s microreactor technology as the preferred nuclear energy source for Tillman’s future AI data center campuses across the United States. The non-binding agreement, announced on August 24, lays out a roadmap to deploy over 2 GW of advanced nuclear capacity by the mid-2030s, scaling to 6 GW by 2040.
What the deal actually includes The framework designates NANO Nuclear’s KRONOS MMR microreactor, rated at 15 megawatts electric, as Tillman Global’s go-to nuclear energy solution. The collaboration spans multiple phases: site evaluation, licensing support, and project development.
On the financial side, the structure is built around milestone-based equity incentives. NANO Nuclear could receive warrants worth up to $100 million in NNE stock, contingent primarily on binding reactor purchase commitments. There’s also an initial restricted stock grant of $5 million baked into the arrangement.
Advertisement
The word “non-binding” deserves emphasis here. Nothing in this framework obligates either party to purchase, build, or deploy anything. The aggressive capacity targets are contingent on definitive agreements and regulatory approval. NANO’s KRONOS MMR is currently at the NRC construction permit pre-application stage, meaning the reactor design hasn’t yet received a construction permit, let alone an operating license.
Why Tillman matters Tillman Global Holdings has amassed approximately $16 billion in capital since its founding in 2013, built on a track record of financing digital infrastructure projects. For NANO Nuclear, a company still in the pre-revenue, pre-construction-permit phase of its reactor development, having a partner with that kind of capital base is significant.
This deal also follows NANO Nuclear’s recent memorandum of understanding with Supermicro for on-site microreactor installations. The pattern is clear: NANO is systematically lining up commercial relationships across the data center ecosystem, building a pipeline of potential customers before the reactor itself is ready for deployment.
The bigger picture: nuclear and the AI power crunch Microreactors represent a particular bet within the nuclear space. Unlike traditional gigawatt-scale nuclear plants, microreactors are designed to be factory-fabricated and deployed at or near the point of consumption. A 15 MWe unit like the KRONOS MMR could theoretically sit adjacent to a data center campus and provide dedicated, carbon-free baseload power without relying on grid transmission infrastructure.
For investors watching NANO Nuclear’s stock, the equity incentive structure in this deal is worth parsing carefully. The $100 million in potential warrants is tied to binding reactor purchases, not to the framework itself. The $5 million restricted stock grant provides more immediate value, but it’s modest relative to the scale of the capacity targets being discussed.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.
Analysis: Bitcoin is in the "extreme discount" zone of the Rainbow Chart, with its downward deviation surpassing that of the previous bear market bottom.
CryptoQuant analyst Axel Adler Jr. noted in a post that Bitcoin is currently trading in the lowest range of the Rainbow Chart model — "basically a fire sale price". This represents an extreme discount relative to the cryptocurrency’s long-term price trajectory, and is not merely a measure of how much Bitcoin has declined. The Rainbow Chart model compares Bitcoin’s current price to its long-term trend, so the gap reflects the market’s deviation from historical patterns. The volatility-adjusted Z-Score now stands at -2.293, the lowest reading since 2016 and lower than the -1.979 recorded at the 2022 bear market bottom. This metric accounts for volatility differences across distinct market cycles, meaning the current downward deviation from the model is more severe than that seen at the last bear market’s trough. Bitcoin is now in an extreme discount zone, with the volatility-adjusted Z-Score hitting a 10-year low. This signals significant undervaluation relative to the model, but it does not alone confirm the final market bottom. Extremely low valuations point to a potentially attractive entry range, though a market reversal would need separate confirmation. Note: The Z-Score is a widely used standardized statistical metric that quantifies how far a data point lies from the mean, measured in units of standard deviation.
42 minutes ago
Stepping into the fray! DeFiLlama reported the fraudulent app for months to no avail, and only after testing it themselves and falling victim to theft did Apple finally take it down from its App Store.
DeFiLlama founder 0xngmi posted on social media that for months, he has been working to get Apple’s App Store to take down a fake app impersonating DeFiLlama, repeatedly reporting trademark infringement and unauthorized impersonation of the official app to Apple. The team then loaded a small amount of funds into a test wallet, downloaded the fake application, and as expected, all funds in the wallet were stolen. The incident was reported to Apple, and the app was removed from the store within days. 0xngmi noted that he hopes other cryptocurrency companies will learn from this to avoid wasting time on similar efforts, as his team did.
42 minutes ago
Iran says it has reached an agreement with Oman on the plan for passage through the Strait of Hormuz.
According to CCTV News, on local time August 15, Iranian Foreign Ministry spokesperson Baghaei said that despite US obstruction, talks between Iran and Oman are still advancing actively, and the two sides have reached an agreement on a navigation plan for the Strait of Hormuz.
42 minutes ago
Thrive Capital took a $215 million position in Amazon shares.
According to Bloomberg, Joshua Kushner’s Thrive Capital has purchased approximately $215 million worth of Amazon shares, continuing the venture capital firm’s strategy of expanding into public market investments. The investment was disclosed in a regulatory filing on Friday, giving Thrive exposure to a large company that stands to benefit from artificial intelligence. Amazon gains from AI through offerings such as generative AI shopping tools and cloud computing infrastructure for other businesses. Earlier this month, Amazon’s market capitalization crossed $3 trillion for the first time, making it the fifth company to reach that milestone. Thrive declined to comment, while a representative for Amazon did not immediately respond to a request for comment. Thrive is known for early investments in firms including SpaceX, Stripe and OpenAI—OpenAI also has backing from Amazon. The VC firm’s other public market holdings include Figma Inc., StubHub Holdings Inc., and Oscar Health, which Kushner co-founded and incubated.
42 minutes ago
Wedbush Reiterates SanDisk's 'Outperform' Rating, Sets Target Price at $2,000
Wedbush reaffirmed its "Outperform" rating on SanDisk stock, setting a target price of $2,000. Wedbush analyst Matt Bryson stated that the firm believes its 2028 projections likely underestimate SanDisk's profitability. The company will repurchase a large volume of shares over the coming years; coupled with extended transaction cycles, these factors warrant a certain valuation premium.
42 minutes ago
Unitree Robotics IPO: Grey market trading rumors are unfounded, and the underlying risks should not be underestimated.
According to a report by Shanghai Securities News, recent market discussions have claimed that scalpers are buying shares of Unitree Technology off-exchange, with some intermediaries offering prices far exceeding the IPO offering price. In response, reporters verified with multiple senior market participants including investment institutions, private equity firms, and securities brokerages, all of whom stated they have not observed such activities, adding that the relevant rumors are largely unfounded. They also noted that such private off-exchange agreements and transactions are of questionable compliance, and investors are highly likely to suffer losses. CITIC Securities, the lead underwriter for Unitree Technology's IPO, also reminded that securities trading must strictly abide by laws, regulations and stock market trading rules, and should only be conducted through legally recognized channels with real-name accounts. It advised investors against blind follow-up speculation, urging them to adhere to rational and value-based investment principles, and carefully assess investment risks.
Not financial or tax advice. PANews content is strictly educational and informational and is not investment advice, financial advice, tax advice, legal advice, or a solicitation to buy or sell any digital asset, security, or financial product. Do your own research and consult qualified advisers.
Disclosure. PANews may publish sponsored content, partner content, advertisements, affiliate links, event promotions, and market commentary involving Web3 projects, service providers, or financial products. PANews personnel, contributors, or affiliates may hold digital assets or other interests related to covered topics. See our Terms of Service.
PlanB: Bitcoin has started bottoming out, a process that typically takes 1 to 3 months.
Renowned crypto analyst PlanB posted that Bitcoin closed at $62,818 in July, with its 200-week moving average standing at $63,000. It appears Bitcoin has entered a bottoming process, which typically takes 1 to 3 months, during which BTC could dip to lower levels.
21 minutes ago
Meme coin ASTEROID's market cap drops below $3 million, falling more than 50% in 20 minutes.
According to GMGN market data, the meme coin ASTEROID on BNB Chain has seen its market capitalization fall below $3 million, with a 52.59% decline in 20 minutes. In an earlier report, BNB Chain officials stated that a former employee had issued the new meme token ASTEROID without authorization using the original official test wallet, and legal action has been taken.
21 minutes ago
BNB Chain: Former Employee Unauthorized to Issue New Meme Tokens via Original Official Test Wallet, Legal Action Taken
BNB Chain released an official statement saying a wallet address was previously created by a former employee, who used it to issue a meme token named TST as part of video tutorials. The employee is no longer with the company. After leaving his position, he retained unauthorized access to the associated seed phrase and used it to generate new private keys. BNB Chain has now learned that the same address is being independently used for activities related to a new meme token, ASTEROID, on BNB Chain. BNB Chain did not create, authorize, promote or participate in the issuance of this token, nor does it have any control over the token or the wallet address. These matters are unrelated to BNB Chain and have not been endorsed by the platform. BNB Chain is taking legal action against the former employee and cooperating with relevant government agencies regarding the incident. Earlier reports noted that the BSC ecosystem meme token ASTEROID hit a $10 million market cap in just four hours after its launch.
21 minutes ago
Morgan Stanley: Semiconductor boom cycle far from over, cloud capital expenditure could surge to $1.3 trillion by 2027
Morgan Stanley’s latest Greater China semiconductor report, released on July 31, notes that AI semiconductors remain in a high-growth cycle, and the current market rally is expanding beyond pure GPU demand to cover advanced process technology, advanced packaging, memory, testing equipment, ASICs, and China’s AI chip industry chain. Morgan Stanley makes a very aggressive forecast: the global cloud AI semiconductor market could reach $485 billion by 2026, and further expand to around $753 billion by 2030. Meanwhile, the global semiconductor market is projected to hit $1.5 trillion by 2030, meaning AI semiconductors will account for nearly half of the total market size. Using its proprietary cloud capital expenditure (capex) tracking model, Morgan Stanley projects that the world’s top 14 listed cloud service providers may see their cloud capex reach nearly $1.3 trillion in 2027, a figure that excludes sovereign AI projects.
21 minutes ago
Analyst: The "fuel" for Bitcoin's forced selling has been exhausted, and investors should watch for signals of capital inflows into Bitcoin ETFs.
Institutions Preview Fed Interest Rate Decision: A hawkish hold could boost the US dollar, with the market focusing on Waller’s remarks.
The Federal Reserve’s interest rate decision will be released tonight, with multiple institutions sharing their forward-looking views. Most institutions expect the Fed to keep interest rates unchanged, and note that the policy statement and Chair Powell’s press conference will be key catalysts for market volatility. Among them: - MUFG and Saxo Bank believe that if the Fed delivers hawkish signals or unexpectedly raises interest rates, the U.S. dollar and U.S. Treasury yields are poised to strengthen. - TD Securities said that if there are no dissenting votes in the decision, it indicates Chair Powell has built internal consensus to some extent, and the U.S. dollar may face a sharp sell-off. - DBS Bank pointed out that even minor changes in the statement’s wording could significantly drive volatility in short-term interest rates. - Goldman Sachs and HSBC argue that even if the Fed holds rates steady throughout the year, the drag on the U.S. dollar will be limited. Unless the Fed unexpectedly raises rates, the dollar lacks new upside catalysts. For gold, ING Group and SIA Wealth Management hold that gold prices still have short-term support, but upside potential may be capped if a hawkish surprise emerges.
9 minutes ago
Key Highlights of the Interest Rate Meeting: Market Focuses on Rate Hike Decision and Number of Hawkish Dissenting Votes
The U.S. Federal Reserve will release its July interest rate decision, with market consensus widely expecting the central bank to hold rates steady. However, potential surprise rate hikes, the policy statement’s wording, and voting splits will be the key highlights of the meeting. Markets will closely watch whether the statement includes more hawkish language, its assessment of inflation risks, and Fed Chair Jerome Powell’s latest views on inflation, the Middle East situation, and future policy paths during his press conference. Institutions widely expect Harker and Logan to cast dissenting votes against keeping rates unchanged, with two dissents already the market’s baseline expectation. If the number of dissenting votes exceeds two, it will be seen as a further sign of hawkish tilt within the Fed, lifting expectations for a September rate hike. Prediction market Kalshi shows Harker and Logan each have a 58% chance of casting a dissenting vote; Polymarket, meanwhile, puts the probability of two dissents at the highest 40%, followed by 21% for one dissent and 13% for three dissents. Market participants believe the meeting’s outcome and Powell’s remarks will drive repricing of the U.S. dollar, U.S. Treasury yields, and gold prices.
9 minutes ago
Ahead of the FOMC statement, the probability of the Federal Reserve raising interest rates this meeting stands at 29.4%.
According to CME’s FedWatch tool, the probability that the Federal Reserve will keep interest rates unchanged today is 70.6%, while the probability of a cumulative 25 basis point rate hike is 29.4%. As for the Fed’s policy path by September, the probability of keeping rates unchanged stands at 17.8%, the chance of a cumulative 25 basis point hike is 60.2%, and the probability of a cumulative 50 basis point hike is 22%.
9 minutes ago
Mounting pressure is building for the Federal Reserve to raise interest rates, and Trump may not necessarily direct his criticism at Walsh.
Foreign media analysis indicates that U.S. President Donald Trump has been pressuring newly appointed Federal Reserve Chair Walsh to cut interest rates as soon as possible, but Wall Street investors are increasingly unanimously betting on the opposite outcome. The recent renewed escalation of tensions in Iran, the implementation of a new round of global tariffs, the ongoing data center investment boom, and sustained strong U.S. consumer spending have together heightened market and Federal Reserve concerns about inflationary pressures, reinforcing expectations that the central bank will maintain tight policy or even raise rates further. Markets widely expect the Fed to hold interest rates steady at its Wednesday meeting. However, whether Walsh, who took over as Fed chair at the end of May, can continue to suppress calls for rate hikes within the committee increasingly depends on whether inflation can improve sustainably. Current polls show U.S. citizens are dissatisfied with Trump’s economic performance, and higher interest rates will undoubtedly further undermine the economic outcomes the White House seeks. Trump has long advocated for interest rate cuts and reiterated this position again this week. Even if the Fed ultimately opts to raise rates, Trump may not initially target Walsh, and is more likely to direct criticism at other Fed officials. Trump has so far appointed three members of the Federal Reserve’s seven-member Board of Governors. He previously said: “Kevin is very capable, but he has a committee, and its members are all highly politicized. He wants to do the right thing, I know what he’s aiming for, but he needs the approval of some people who may have ulterior motives. Interest rates should be lowered.”
9 minutes ago
Bitfinex: Bitcoin long-term holders continue to accumulate, with the market closer to stabilization than a recovery.
Bitfinex stated in a report that despite Bitcoin’s price falling from $82,000 to below $58,000, long-term holders have maintained net accumulation of BTC. However, alongside a concurrent weakening of selling pressure, their accumulation pace has slowed notably: net additions hit roughly 40,000 BTC in late May, dropping to around 14,000 BTC by late July. Bitfinex views the current market as being in a stabilization phase rather than having entered a substantive recovery.
Not financial or tax advice. PANews content is strictly educational and informational and is not investment advice, financial advice, tax advice, legal advice, or a solicitation to buy or sell any digital asset, security, or financial product. Do your own research and consult qualified advisers.
Disclosure. PANews may publish sponsored content, partner content, advertisements, affiliate links, event promotions, and market commentary involving Web3 projects, service providers, or financial products. PANews personnel, contributors, or affiliates may hold digital assets or other interests related to covered topics. See our Terms of Service.
Not financial or tax advice. PANews content is strictly educational and informational and is not investment advice, financial advice, tax advice, legal advice, or a solicitation to buy or sell any digital asset, security, or financial product. Do your own research and consult qualified advisers.
Disclosure. PANews may publish sponsored content, partner content, advertisements, affiliate links, event promotions, and market commentary involving Web3 projects, service providers, or financial products. PANews personnel, contributors, or affiliates may hold digital assets or other interests related to covered topics. See our Terms of Service.
According to public market data, Kong Jianping, founder and chairman of Nano Labs, indirectly holds approximately 18.982 million shares of Changxin Technology via Yifang Changda Fund. Kong’s subscribed capital contribution to the fund is 21.34 million yuan. Calculated based on Changxin Technology’s opening price of 49.5 yuan per share today, the market value of his indirect holdings is roughly 940 million yuan, about 44 times his subscribed capital contribution.
Relevant content
Microsoft AI launches a new self-developed cybersecurity model.
Microsoft AI has launched a new self-developed cybersecurity model, MAI-Cyber-1-Flash. The model operates on the MDASH platform, a multi-agent framework designed to detect and remediate vulnerabilities in large-scale codebases. MDASH announced that the combination of MAI-Cyber-1-Flash and GPT-5.4 scored 12 points higher than Mythos in the CyberGym benchmark test. Microsoft AI stated that the model configuration provided on MDASH is 50% cheaper than its previous optimal setup (GPT-5.4 + 5.4 mini + 5.3 Codex). Source: Jinshi
32 minutes ago
A U.S. organization projects that AI will discover twice as many cybersecurity vulnerabilities this year as it did last year.
The number of software security vulnerabilities found in popular tech products in 2026 is projected to be roughly double the 2025 count, with this surge largely attributed to the growing capabilities of artificial intelligence (AI) systems. The U.S. National Vulnerability Database (NVD) shows that 45,207 vulnerabilities have been logged between January and this Monday, a figure nearly matching the total for all of 2025. Last year, the database recorded an all-time high in vulnerabilities. Oracle (ORCL.N) reported that its July monthly software update fixed 1,449 security vulnerabilities, an all-time high, compared to just 309 fixes in the same period last year. Microsoft (MSFT.O) disclosed 642 security vulnerabilities in July, also an all-time high, nearly five times the number from the same period a year ago. Google (GOOG.O) discovered and fixed 433 such vulnerabilities in its latest Chrome browser update, versus only 11 in the equivalent update a year prior. Gabriel Shapiro, a distinguished AI research scientist at cybersecurity firm SentinelOne, noted: “We must face the reality that these tools are boosting people’s ability to uncover software vulnerabilities.” Doug Turner, Google’s Chrome engineering director, said vulnerabilities are being found at an “unprecedented scale and speed” due to advances in AI models and corresponding investments.
32 minutes ago
Apple faces a lawsuit over a crypto wallet scam on the App Store, with users losing more than $1.8 million.
Apple is being sued by the U.S. District Court for the Northern District of California over allegations it failed to effectively prevent cryptocurrency scam apps on the App Store. Three users allege Apple’s security review mechanisms were flawed, leading them to download fake crypto wallet apps that resulted in total losses exceeding $1.8 million. The lawsuit claims the apps in question impersonate the Bitcoin wallet "Sparrow Wallet", while the official Sparrow Bitcoin Wallet is not actually available on iOS. The three users suffered asset losses after transferring Bitcoin to the fake apps: one lost roughly $875,000, another around $840,000, and the third about $120,000. Plaintiffs argue Apple has long marketed its strict App Store review process as a security advantage, claiming its closed ecosystem reduces malware and scam risks, and thus should be held liable for fraudulent apps on the platform. The lawsuit also cites public criticism from the founder of Sparrow Bitcoin Wallet, stating Apple previously allowed imposter apps to appear on the App Store. The three plaintiffs are seeking a jury trial, restitution of their losses plus additional damages, and demanding Apple enhance risk warnings and security disclosures on the App Store. Apple responded that impersonating other apps violates App Store rules, and the company quickly removes such apps, adding that no imposter Sparrow Wallet apps currently exist on the App Store. Apple previously released data showing its review team rejected over 371,000 app submissions in 2025 that involved impersonation, spam, or user deception.
32 minutes ago
Bitcoin pullback hits crypto treasury firms: TD Cowen slashes Nakamoto’s target price by 58% while retaining a Buy rating.
Wall Street investment bank TD Cowen has cut the price target for Bitcoin treasury company Nakamoto Inc. (NASDAQ: NAKA), slashing the post-stock-split adjusted target from $40 to $17—a 58% reduction—while retaining its "Buy" rating. TD Cowen analysts said the adjustment is mainly driven by pressure from Bitcoin price declines on Nakamoto’s highly leveraged capital structure. While the new target still implies around 275% upside from the current share price of $4.65, the stock is highly sensitive to Bitcoin price swings. TD Cowen forecasts Bitcoin will rebound to $100,000 by the end of 2026, roughly 25% below its all-time high of $126,000 set last October. The firm also expects Nakamoto to pause further Bitcoin purchases before 2027. Analysts noted that Nakamoto’s core value still stems from its Bitcoin holdings: the company currently holds 4,467 BTC worth approximately $290 million, ranking 22nd among public companies globally in Bitcoin holdings. However, its debt and preferred stock financing structure has eroded the asset value available to common shareholders. Recently, Nakamoto has completed several financial adjustments, including repaying roughly $45 million in debt, extending the maturity of $105 million in principal to June 2027, reducing financing costs, and approving a $25 million share repurchase program. Additionally, the company has shut down its previously operated medical clinic business and will focus on Bitcoin media, asset management, and advisory services going forward. Data shows NAKA’s share price has fallen more than 71% year-to-date, while Bitcoin has dropped around 26% over the same period. Market attention is shifting from "continuous BTC purchases" to the balance sheet structure and financing capabilities of Bitcoin treasury companies.
32 minutes ago
Registrations for Trump’s account top 7 million, U.S. Treasury Secretary hails it as "the most successful launch project in government history"
U.S. Treasury Secretary Scott Bessent has dubbed Trump Accounts "the most successful launch project in U.S. government history," disclosing that roughly 7 million children have already registered for the initiative. Bessent made the remarks at a meeting of the Financial Literacy and Education Commission, noting that the number of registrants has risen from 6.5 million earlier this month. The U.S. Treasury Department added that the early sign-up rate for Trump Accounts outpaces that of other digital platforms and financial products. Launched on July 4, the program – also known as 530A accounts – is open to all U.S. children under 18 who hold a Social Security number. Under the plan, children born between 2025 and 2028 will receive a one-time $1,000 initial deposit from the U.S. Treasury. Parents, guardians, and grandparents can contribute up to $5,000 annually, with funds invested in ETFs tracking the S&P 500 index. Bessent stated that the project will help ordinary U.S. families access capital markets, "creating a new generation of shareholders" and enabling households long excluded from Wall Street to participate in stock investing. Consultancy McKinsey previously analyzed that with broad participation, Trump Accounts could accumulate approximately $80 billion to over $900 billion in assets for U.S. children over the next decade, though the final scale will depend on family participation rates, willingness to make ongoing contributions, and long-term investment performance.
32 minutes ago
NVIDIA and OpenAI are developing a planned $500 billion AI data center project, with a 10GW capacity that could make it the world's largest.
According to The Wall Street Journal (WSJ), citing people familiar with the matter, NVIDIA and OpenAI are in talks for a mega AI data center project worth approximately $500 billion, located in southern Ohio, United States. The project is expected to have a maximum power capacity of 10GW, potentially becoming the world’s largest data center initiative. Reports state that NVIDIA may provide around $250 billion in funding to OpenAI for leasing a data center campus developed by SBEnergy, the energy subsidiary of SoftBank. The total cost of the project is projected to exceed $500 billion, including up to $350 billion for AI chip procurement. The first phase of the data center is scheduled for completion in 2028, with an initial power capacity of roughly 800MW. If fully implemented, its 10GW power scale will far outpace most current large-scale AI infrastructure projects, equivalent to meeting the electricity needs of around 8.4 million U.S. households. The project’s power resources are located on U.S. federal land and are jointly supported by the U.S. and Japan. Japan previously pledged $33 billion in investments for related energy infrastructure in exchange for reduced tariff arrangements. As demand for AI computing power continues to surge, NVIDIA is accelerating its expansion into large-scale AI infrastructure, forging deep partnerships with companies including OpenAI, Meta, Microsoft, and Amazon. If the project moves forward, it will further solidify NVIDIA’s core position in the AI computing power supply chain and push global AI infrastructure into a "trillion-dollar-level" competitive phase.
NVIDIA: Major Clients Have Begun Testing Vera Rubin Devices
According to Bloomberg, NVIDIA has announced that its key clients have started testing its Vera Rubin devices. The chipmaker added that its new Vera processor outperforms AMD’s Turin, and that the chips are being delivered on schedule for use in AI data centers.
3 minutes ago
Iran's Revolutionary Guard hits U.S. military radar in Kuwait.
According to Iran's Press TV, Iran's Revolutionary Guard hit a U.S. military radar at Kuwait's Al Jaber Base.
3 minutes ago
Liang Wenfeng’s Huanfang and Jiuzhang secure the largest share in Changxin Technology’s private placement new share offering, with 113 private equity firms receiving allocations.
The preliminary offline placement results for Changxin Technology show that a total of 2,459 products under 113 private equity firms secured offline placements in the company, with a total of 161 million shares allocated, amounting to 1.436 billion yuan. The announcement notes that offline institutional investors are divided into Category A (public funds, social security funds, pension funds, enterprise annuities, bank wealth management products, insurance companies, QFIIs) and Category B (private equity firms, broker-dealer proprietary trading, trusts, financial companies, etc.). Category A investors, dominated by public funds, received 1.978 billion shares, accounting for 91% of the total offline issuance; while Category B investors, led by private equity firms, secured 196 million shares, making up only 9% of the total offline issuance. Among the private equity placement list, the top ten by number of placement objects are all leading quantitative private equity firms. Shanghai Yanfu has a total of 282 placement objects allocated, ranking first among private equity firms; Century Front, Jiukun Investment, Shanghai Chengqi, and Huanfang Quant have 209, 194, 167, and 153 placement objects respectively; Lingjun Investment, Shanghai Jinde, and Minghong Investment also have over 100 allocated products each, at 107, 105, and 100 respectively. Notably, Liang Wenfeng, founder of DeepSeek and a prominent private equity figure, took the largest share among private equity placements. Public information shows that the actual controllers of two leading 100-billion-yuan private equity firms, Ningbo Huanfang Quant and Zhejiang Jiuzhang Asset Management, are both Liang Wenfeng. This means that through his two private equity firms, Liang Wenfeng has a total of 194 private equity products allocated, with a total of 20.2497 million shares secured, amounting to approximately 175 million yuan. (The Paper)
3 minutes ago
Pump.fun launches BOOST mode, aiming to re-inject permanently locked liquidity into the token market.
Meme coin launch platform pump.fun has announced the launch of its new BOOST mode, set as the default launch mechanism for all new Pump.fun tokens moving forward. The feature is designed to address the long-standing "dead liquidity" problem during token migrations, using a buyback and burn mechanism to re-inject liquidity that was previously permanently locked back into the token market. Pump.fun noted that over $100 million in liquidity is permanently lost annually during token migrations, with these funds no longer available to support market liquidity. Historically, roughly 20% of liquidity remains stuck in liquidity pools (LPs) for every token that completes migration — even after all traders sell their positions, some funds stay locked in the pools permanently. BOOST mode will leverage this trapped liquidity to re-inject into the market via an automatic buyback mechanism within 5 minutes of each token migration completion. Specifically, BOOST will execute buybacks using a post-migration time-weighted average price (TWAP) and automatically burn the purchased tokens. For SOL trading pairs, 17.6 SOL will be injected, while USDC trading pairs will receive $2,516 in funds. The mechanism requires no manual activation from users: all new Pump.fun tokens that complete migration after 10:23 AM Eastern Time (ET) on July 21 will automatically enable the BOOST configuration. Tokens migrated prior to this date or issued via the Mayhem platform do not include the feature. The upgrade aims to improve trading experiences and enhance the long-term utilization efficiency of liquidity within the ecosystem.
3 minutes ago
Telegram Founder: Will Integrate a Native Non-Custodial Gram Wallet for All Users
Telegram founder Pavel Durov announced in his personal channel that instant, zero-fee cryptocurrency transactions for its more than 1 billion users are set to become a reality. The platform is adding a native, non-custodial Gram wallet to every Telegram application.
3 minutes ago
GRAM surges past $1.5, gaining over 9% in 10 minutes.
According to HTX market data, GRAM has broken through $1.5, currently trading at $1.555, up over 9% in 10 minutes. Earlier reports stated that Telegram’s founder said the team is building native non-custodial Gram wallets into every Telegram application.
NVIDIA: Major Clients Have Begun Testing Vera Rubin Devices
According to Bloomberg, NVIDIA has announced that its key clients have started testing its Vera Rubin devices. The chipmaker added that its new Vera processor outperforms AMD’s Turin, and that the chips are being delivered on schedule for use in AI data centers.
3 minutes ago
Iran's Revolutionary Guard hits U.S. military radar in Kuwait.
According to Iran's Press TV, Iran's Revolutionary Guard hit a U.S. military radar at Kuwait's Al Jaber Base.
3 minutes ago
Liang Wenfeng’s Huanfang and Jiuzhang secure the largest share in Changxin Technology’s private placement new share offering, with 113 private equity firms receiving allocations.
The preliminary offline placement results for Changxin Technology show that a total of 2,459 products under 113 private equity firms secured offline placements in the company, with a total of 161 million shares allocated, amounting to 1.436 billion yuan. The announcement notes that offline institutional investors are divided into Category A (public funds, social security funds, pension funds, enterprise annuities, bank wealth management products, insurance companies, QFIIs) and Category B (private equity firms, broker-dealer proprietary trading, trusts, financial companies, etc.). Category A investors, dominated by public funds, received 1.978 billion shares, accounting for 91% of the total offline issuance; while Category B investors, led by private equity firms, secured 196 million shares, making up only 9% of the total offline issuance. Among the private equity placement list, the top ten by number of placement objects are all leading quantitative private equity firms. Shanghai Yanfu has a total of 282 placement objects allocated, ranking first among private equity firms; Century Front, Jiukun Investment, Shanghai Chengqi, and Huanfang Quant have 209, 194, 167, and 153 placement objects respectively; Lingjun Investment, Shanghai Jinde, and Minghong Investment also have over 100 allocated products each, at 107, 105, and 100 respectively. Notably, Liang Wenfeng, founder of DeepSeek and a prominent private equity figure, took the largest share among private equity placements. Public information shows that the actual controllers of two leading 100-billion-yuan private equity firms, Ningbo Huanfang Quant and Zhejiang Jiuzhang Asset Management, are both Liang Wenfeng. This means that through his two private equity firms, Liang Wenfeng has a total of 194 private equity products allocated, with a total of 20.2497 million shares secured, amounting to approximately 175 million yuan. (The Paper)
3 minutes ago
Pump.fun launches BOOST mode, aiming to re-inject permanently locked liquidity into the token market.
Meme coin launch platform pump.fun has announced the launch of its new BOOST mode, set as the default launch mechanism for all new Pump.fun tokens moving forward. The feature is designed to address the long-standing "dead liquidity" problem during token migrations, using a buyback and burn mechanism to re-inject liquidity that was previously permanently locked back into the token market. Pump.fun noted that over $100 million in liquidity is permanently lost annually during token migrations, with these funds no longer available to support market liquidity. Historically, roughly 20% of liquidity remains stuck in liquidity pools (LPs) for every token that completes migration — even after all traders sell their positions, some funds stay locked in the pools permanently. BOOST mode will leverage this trapped liquidity to re-inject into the market via an automatic buyback mechanism within 5 minutes of each token migration completion. Specifically, BOOST will execute buybacks using a post-migration time-weighted average price (TWAP) and automatically burn the purchased tokens. For SOL trading pairs, 17.6 SOL will be injected, while USDC trading pairs will receive $2,516 in funds. The mechanism requires no manual activation from users: all new Pump.fun tokens that complete migration after 10:23 AM Eastern Time (ET) on July 21 will automatically enable the BOOST configuration. Tokens migrated prior to this date or issued via the Mayhem platform do not include the feature. The upgrade aims to improve trading experiences and enhance the long-term utilization efficiency of liquidity within the ecosystem.
3 minutes ago
Telegram Founder: Will Integrate a Native Non-Custodial Gram Wallet for All Users
Telegram founder Pavel Durov announced in his personal channel that instant, zero-fee cryptocurrency transactions for its more than 1 billion users are set to become a reality. The platform is adding a native, non-custodial Gram wallet to every Telegram application.
3 minutes ago
GRAM surges past $1.5, gaining over 9% in 10 minutes.
According to HTX market data, GRAM has broken through $1.5, currently trading at $1.555, up over 9% in 10 minutes. Earlier reports stated that Telegram’s founder said the team is building native non-custodial Gram wallets into every Telegram application.
The Ledger Nano X remains one of the best-selling hardware wallets on the market, and at $149 through Ledger’s official store it sits squarely between the entry-level Nano S Plus ($79) and the premium touchscreen Ledger Stax ($399). Launched in 2019 as a mobile-friendly upgrade to the original Nano S, it’s built around a CC EAL5+ certified Secure Element chip, pairs over Bluetooth with iOS and Android, and supports 5,500+ coins and tokens through Ledger Live and connected third-party wallets. Ledger Nano X reviews online tend to agree on the fundamentals but split on value for money — this hardware wallet review breaks down what the device actually gets you for the price, where it falls short, and who should buy the cheaper Nano S Plus instead.
Key Takeaways The Ledger Nano X costs $149 and is Ledger’s mid-range hardware wallet, positioned between the Nano S Plus ($79) and the Ledger Stax ($399). It’s the only Ledger device that combines Bluetooth connectivity with a built-in battery, letting it pair directly with iOS and Android through Ledger Live. Ledger advertises support for 5,500+ coins and tokens, though native Ledger Live app support is narrower — the larger figure includes coins reachable only through third-party wallets like MetaMask. Security is built around a CC EAL5+ certified Secure Element chip; the chip itself has never been compromised, though a 2020 customer-data leak and the optional 2023 Ledger Recover service remain points of criticism. The Nano S Plus is the better buy for anyone who doesn’t specifically need Bluetooth or mobile pairing. What Is the Ledger Nano X? The Ledger Nano X is a cold-storage hardware wallet — a small physical device, roughly the size of a USB stick, that stores the private keys controlling access to a user’s cryptocurrency. Those keys never leave the device: transactions are signed on the Nano X itself, so the wallet stays secure even when connected to a compromised computer or phone. Ledger launched the Nano X in 2019 as a premium successor to the original Nano S, and it has remained one of the company’s flagship models since, thanks to broad coin support and Bluetooth-based mobile use.
Key Specs at a Glance SpecDetailPrice$149Security chipCC EAL5+ certified Secure Element (ST33K1M5)Coin support5,500+ coins/tokens across 100+ chains via Ledger Live and connected walletsConnectivityBluetooth + USB-CBatteryUp to ~8 hours standbyApp storageUp to 100 apps (real-world capacity is smaller for larger apps like Bitcoin or Ethereum)DimensionsApprox. 72mm x 18.6mm x 11.75mm, 34g Security The Secure Element Chip The Nano X’s private keys are generated and stored on a CC EAL5+ certified Secure Element chip — the same class of certification used in passports and payment cards. This chip has not been compromised in Ledger’s history, and the design means keys never touch an internet-connected device during a transaction.
The Ledger Recover Controversy In 2023, Ledger introduced an optional subscription service called Ledger Recover, which allows users to split and back up an encrypted version of their recovery phrase across third-party custodians. It’s opt-in and disabled by default, but its existence changed the trust model some long-time users expected from a fully offline device, and it continues to draw criticism from parts of the Bitcoin community.
The 2020 Data Leak In July 2020, Ledger disclosed that roughly 270,000 customer records — names, emails, phone numbers, and shipping addresses — were exfiltrated from a marketing database. No funds or private keys were affected, but the leak fuels phishing campaigns against Ledger customers to this day. Anyone using a Ledger device should treat unsolicited emails or calls claiming to be from Ledger with default suspicion.
Setup and Ledger Live Setting up the Nano X follows Ledger’s standard flow: unbox the device, initialize it, set a PIN code, write down the 24-word recovery phrase on the included cards, then install the Ledger Live app on desktop or mobile to add coin apps. Ledger Live also supports buying, swapping, and staking assets directly, though swap fees through Ledger Live tend to run higher than going through a dedicated DEX aggregator.
Bluetooth and Mobile Use The standout feature separating the Nano X from the cheaper Nano S Plus is Bluetooth. The Nano X pairs with iOS and Android through Ledger Live, letting users check balances, sign transactions, and install apps without a cable. It’s genuinely useful for anyone who manages crypto primarily from a phone, though some users report occasional Bluetooth pairing issues — worth factoring in if wireless reliability matters more to you than the convenience itself.
Pros and Cons Pros:
Broad coin support across major chains and thousands of tokens Only Ledger device with both Bluetooth and a battery for mobile use Chip-level security with no history of compromise Works with major third-party wallets, including MetaMask, Phantom, and Rabby Cons:
Real-world app storage is smaller than the “100 apps” figure suggests once larger coin apps are installed $149 is a meaningful step up from the $79 Nano S Plus for buyers who don’t need Bluetooth Closed-source firmware Setup can feel clunky for first-time hardware wallet users Ledger Nano X vs. Nano S Plus vs. Trezor Model T Ledger Nano XLedger Nano S PlusTrezor Model TPrice$149$79$219BluetoothYesNoNoScreenSmall monochromeSmall monochromeColor touchscreenCoin support5,500+5,500+1,800+ The Nano S Plus supports essentially the same coin range as the Nano X for nearly half the price — the Nano X’s premium is almost entirely the Bluetooth and battery. The Trezor Model T costs more and adds a touchscreen but supports fewer coins natively.
Who Should Buy the Ledger Nano X? The Nano X makes the most sense for holders with diversified, multi-chain portfolios who want to manage crypto from a phone as often as a desktop. Active Solana, Ethereum, or multi-chain traders who pair a Ledger with software wallets like Phantom or MetaMask for daily use get the most value from the mobile flexibility. Anyone holding one or two coins long-term and rarely needing mobile access is better served by the cheaper Nano S Plus. For readers still deciding between cold storage and a software wallet, our Trust Wallet review breaks down the hot-wallet side of that trade-off.
Bottom line for this Ledger review: the Nano X earns its higher price through Bluetooth and mobile flexibility, not through better security or wider coin support than the Nano S Plus — both run the same certified chip.
This article is for informational purposes only and does not constitute financial advice. Always conduct independent research before purchasing hardware or managing your own crypto security.
Frequently Asked Questions Is the Ledger Nano X safe? Yes. Its CC EAL5+ certified Secure Element chip has never been compromised, and private keys are generated and stored on the device, never touching an internet-connected computer or phone during a transaction. The real risks are user-side rather than hardware-side — phishing attempts tied to Ledger's 2020 customer-data leak remain the most common threat, not any flaw in the device's chip security.
How many coins does the Ledger Nano X actually support? Ledger advertises support for 5,500+ coins and tokens, but that figure includes assets reachable only by connecting the device to third-party software wallets like MetaMask, Phantom, or Rabby. Native support inside the Ledger Live app alone is narrower, covering major chains like Bitcoin, Ethereum, and Solana directly, with broader altcoin access requiring an external wallet.
Ledger Nano X vs Nano S Plus — which should you buy? Both devices run the same certified Secure Element chip and support essentially the same range of coins, so security isn't the differentiator. The Nano X costs $70 more but adds Bluetooth connectivity and a built-in battery for managing crypto from a phone. If you mainly use a desktop and don't need mobile access, the $79 Nano S Plus offers the same protection for less.
Has the Ledger Nano X ever been hacked? The Secure Element chip itself has no history of being compromised, and no Ledger device has had funds stolen directly through a chip-level exploit. Ledger's one major security incident was a 2020 leak of customer contact information — names, emails, and addresses — from a marketing database, which did not expose private keys or wallet funds but has fueled ongoing phishing campaigns.
Is the Ledger Nano X worth $149? For users who hold a diversified, multi-chain portfolio and want to manage it from a phone as often as a desktop, the Bluetooth and mobile flexibility justify the price. For users who only need to secure one or two assets long-term and rarely need mobile access, the $79 Nano S Plus provides nearly identical security for significantly less money.
AUTHOR
Mushumir Butt is a seasoned crypto journalist with over three years of experience reporting on the world of blockchain and cryptocurrency. At Blockchain Reporter, he delivers insightful news, in‐depth project reviews, and precise price analysis and predictions. With a strong background in SEO and digital marketing, Mushumir excels at breaking down complex trends into clear, accessible content, ensuring readers stay ahead in the fast‐paced crypto space.
In brief Nano Banana 2 Lite (gemini-3.1-flash-lite-image) generates images in four seconds at roughly $0.034 per image. This means it produces results at about half the cost of Nano Banana 2 at the same resolution and 2.7× faster. In head-to-head testing, the Lite model matched or beat Nano Banana 2 on many fields, but when details are important, the more expensive version may be the better option. Google last week launched Nano Banana 2 Lite—officially gemini-3.1-flash-lite-image—as the entry point in its image generation stack, sitting below Nano Banana 2 and well below Nano Banana Pro. It delivers text-to-image outputs in roughly four seconds, 2.7 times faster than Nano Banana 2, and is positioned as the direct replacement for the original Nano Banana (gemini-2.5-flash-image). The explicit pitch: same Google ecosystem, less money, less waiting.
The model is available through Google AI Studio, the Gemini API, and the Enterprise Agent Platform—and it's baked into consumer products including Search, the Gemini app, NotebookLM, and Google Photos. It works alongside Gemini Omni Flash, Google's new video generation model, through the Interactions API, which lets users stack up to three sequential edits within a single session. The Nano Banana family now reads as a clean three-tier structure: Lite for speed and cost, Nano Banana 2 for the quality-speed balance, Nano Banana Pro for complex professional work.
At roughly $0.034 per image at 1K resolution, Nano Banana 2 Lite is about half the price of Nano Banana 2, which runs $0.067 per image at the same resolution. That puts the Lite model in direct competition with Seedream 5.0 Lite, which comes in at $0.031–0.035 per image. Reve 2.0 undercuts both at around $0.0067 per image via API—though it lacks the deployment breadth that comes with Google's infrastructure. Qwen Image Edit is a good, free, open-source option for standard use cases.
So, is the quality drop from Nano Banana 2 concentrated enough to matter for your specific workflow? Is it distributed enough that most people won't notice?
We ran the same prompts through both models across five categories to find out. The answer is less predictable than you'd expect.
Realism
The realism test is where the gap between Nano Banana 2 and its Lite sibling is most visible. Both models received the same technically demanding portrait prompt: a cinematic image of a 32-year-old female architect on a rooftop at sunset, wearing a beige trench coat and round glasses, holding rolled blueprints specifically in her left hand, with a defocused city skyline behind her, golden hour lighting with a soft rim light, shallow depth of field simulating a 50mm lens, a vertical 4:5 aspect ratio, realistic skin texture, and subtle film grain.
The prompt explicitly frames each element as an independent constraint that can fail.
Nano Banana 2 Lite passed the basic test. The subject is correctly dressed and positioned, wears round glasses, holds blueprints, and stands on a rooftop with a blurred city behind her. But it is slightly, just slightly, less realistic in terms of details: The subject only has one hand, which is oversized in comparison to the rest of the body. The rim light is barely perceptible. Skin texture holds up at thumbnail scale but doesn't survive close inspection. The image, in the end, looks like a competent stock photo, not a cinematic portrait.
Nano Banana 2 produced something photographically different in kind. The subject stands against a fully realized New York City skyline at magic hour, bokeh city lights blooming across the background, a hint of a river visible in the distance. The depth of field is dramatic. The warm rim light clearly separates the subject from the background. The blueprints are in her left hand, not her right hand, as requested.
Both models struggle with symmetry. For example the holes for the buttons and some straps are not consistent, but again, those are details that are spotted upon closer inspection.
For social media content or rapid visual mockups, the Lite version is workable—it communicates the concept. For anything where the image is the final product—a hero image, a client deliverable, a portfolio piece—it will show its seams at any resolution above a thumbnail. Photographic quality is where the Lite model's architecture makes its largest single concession, and it makes it consistently.
Prompt Adherence
Prompt adherence testing used a different strategy: a dense, multi-element scene where each labeled detail functions as an independent failure point. The prompt described a steampunk cityscape viewed from a gargoyle's perch—complete with a hot air balloon labeled "Atlas & Sons Cartographers, Est. 1842," a cable car with a specific named route, a gear-driven clock tower, a gargoyle holding a document labeled "Sector 7 – Condemned," a foreground newspaper with a specific headline, and a detailed Victorian street scene below.
The logic: If a model can hold 10 specific simultaneous constraints, you can trust it on complex creative briefs.
Both models produced visually compelling steampunk scenes. Both correctly place the gargoyle in the foreground, the clock tower at center, the balloon in the sky, and a cable car crossing the frame. At a glance, the differences feel cosmetic—the Lite version is darker and moodier, the full model cleaner and brighter. But the specifics tell a different story. In the Lite version, the balloon reads "Est. 1942" instead of 1842—mostly due to AI grappling to properly render text. The cable car route label is partially garbled. The foreground newspaper headline blurs at the edges, losing legibility on the details that were specifically requested.
Overall, it focused more on visuals than text, which is ok for most use cases.
Nano Banana 2 gets almost everything right. The balloon clearly reads "Atlas & Sons Cartographers Est. 1842." The cable car sign says "Upper Vantis – 4 Stops." The gargoyle holds a document, but the text is illegible. The foreground newspaper reads "Clocktower Falls Silent – City Mourns" in clean, readable type. Every named element appears where it should, with the correct label, in legible form. The compositional decision to use brighter, more editorial lighting also pays off here—it keeps the labeled details readable rather than swallowed by atmosphere.
Casual prompt users won't catch a one-digit transposition on a fictional establishment date. But concept artists, worldbuilders, and narrative illustrators—the people using these models to communicate specific creative logic to clients or collaborators—will notice immediately.
The Lite model's tendency to blur or transpose specific in-image text labels isn't a catastrophic failure, but it introduces a manual correction step that compounds badly at scale.
Spatial Awareness
Spatial awareness testing evaluated how each model handles multi-depth scene composition: multiple objects at close range, a human subject in the middle distance, and atmospheric elements receding into background darkness.
The scene—a medieval alchemist at a cluttered wooden desk, surrounded by an armillary sphere, a lit candle, an hourglass, a skull, star charts, and a glowing green jar, with a black cat silhouetted in an arched window behind him—requires convincing three-dimensional layering to read as coherent rather than assembled.
Both models understood the basic spatial grammar of the scene. Foreground objects are rendered at appropriate scale and shadow detail, the scholar occupies the mid-ground with correct occlusion relationships to the objects around him, and the arched window with the moonlit night sky creates a convincing sense of recession behind the scene. Neither model misplaces objects, collapses depth planes, or introduces spatial contradictions. The scene architecture—front, middle, back—is correctly established in both outputs.
The differences are subtle and real. Nano Banana 2's version has a richer atmospheric depth gradient: The candlelight fades naturally as it reaches the stone walls, the background haziness reads as genuine atmospheric depth rather than digital softening, and the overall scene has a painterly warmth that suggests volumetric space. The Lite version's depth is structurally correct but slightly compressed—the background reads marginally more like a stage flat than a receding room with actual air in it.
At least in this text, the Nano Banana 2 image feels like the same Nano Banana 2 Lite image with a detailed LoRA (a sort of specialized fine tuning layer) applied during sampling.
This is the smallest gap across all five tests. For storyboards, game asset concepts, and most editorial illustration contexts, both models demonstrate adequate spatial reasoning. The Lite model's slightly flatter depth rendering becomes meaningful only in high-resolution output or detailed compositional analysis—and even then, the gap is arguable.
For this category, the Lite model is a viable substitute in the vast majority of practical workflows.
Text Generation
Text generation is where this review produces its most counterintuitive result.
The test prompt described a gritty nighttime hardware store with dozens of simultaneous text elements at different scales and styles: a hand-painted main sign with the store name, founding date, and product categories; a graffiti tag on the façade; window decals with hours and services; a concert poster with band name, venue, date, doors time, and specific ticket prices; a city council meeting notice; a lost cat notice with a phone number; political stickers on a phone booth; and a street parking restriction on the curb.
Text generation at this complexity is difficult because each element has to be correctly rendered while the overall image still reads as a coherent photograph.
Nano Banana 2 Lite actually delivered something genuinely impressive for how fast it is. "KELLERMAN'S HARDWARE & SUPPLY CO. – SINCE 1931 – TOOLS, ROPE, PAINT," graffiti reading "STILL HERE," window signs for "OPEN 7 DAYS / WE BUY SCRAP – ASK FOR RAY / CLOSED," a concert poster for "THE DREDGE PALE MOUTH / SUNDAY JUNE 4 / DOORS 9PM / THE ANCHOR CLUB / $12 ADV – $15 DOOR," stickers reading "THIS MACHINE KILLS FASCISTS" and "JESUS SAVES," a lost cat notice with a specific and legible phone number—every single text element in the prompt is correctly rendered and readable simultaneously in one image.
If there’s something to note, it’s that the image is less realistic. Some posters seem rendered by an editor with poor photoshop skills rather than genuine elements of the scene. One example could be the posters pasted on the phone booth. To be more realistic they should have some natural imperfections, and even deterioration signs. That said, this is a legitimately strong result for any image model, let alone the cheaper, faster one.
Nano Banana 2's version is also strong. Most text is correctly placed and legible, and the overall image reads as a convincing nighttime scene. But the full model's darker, moodier atmospheric rendering—generally one of its assets—works against it here. Several smaller sticker texts fall into shadow and lose legibility. The Lite model's brighter, more neutral lighting, a quality that reads as a weakness in portrait work, becomes a clear advantage when the evaluation criterion is whether all the text in the scene is actually readable.
For text-heavy generation—signage mockups, editorial graphics, product concepts with labeled elements, infographic-style composed images—Nano Banana 2 Lite performs below Nano Banana 2. The model seems to either focus too much on visuals that text becomes garble, or focus so much on text that its placement in scene becomes unrealistic.
ConclusionsNano Banana 2 Lite is not a straight downgrade from Nano Banana 2. It's a focused tool with a specific ceiling, and that ceiling drops hardest in exactly the scenarios where photographic quality is the deliverable, and holds surprisingly steady everywhere else.
Cinematic portrait work, sophisticated lighting physics, fine material texture, close-inspection-quality skin rendering—all of these expose a clear difference between the two models. Style transfer also takes a meaningful hit, not in rendering quality but in contextual comprehension: the Lite model can execute a subject, but it struggles to capture the visual environment in which that subject lives. Prompt adherence degrades specifically on in-image labeled text accuracy—a narrow failure mode, but one that matters badly in worldbuilding, concept art, and any pipeline where specific in-image language carries meaning.
What holds up well—and in some cases holds up better—is specificity: if you require a lot of focus on something, it will make sure everything is there.
Spatial scene architecture, and basic compositional competence are also good. The text generation result warrants specific emphasis: If your workflow involves signage mockups, branded graphics, editorial composites with text-heavy elements, or any pipeline where multiple readable text strings need to coexist in a single image, the Lite model is worth reaching for first. Its brighter rendering defaults, a liability in portrait work, are an advantage when legibility is the metric. Spatially, it handles multi-depth scenes adequately for the vast majority of professional contexts.
On the cost math: at $0.034 per image, Nano Banana 2 Lite runs at roughly half the cost of Nano Banana 2 at 1K resolution ($0.067) and trades almost blow-for-blow with Seedream 5.0 Lite ($0.031–0.035). Reve 2.0 undercuts both dramatically at approximately $0.0067 per image via API, but doesn’t offer the deployment footprint that comes with the Nano Banana ecosystem: Search, NotebookLM, Google Photos, and the Gemini app running off the same model simultaneously.
For teams already inside Google's infrastructure, that integration removes a platform-switching cost that pure-API alternatives can't account for. If you know which use cases you're in—and you're not in the photographic quality bucket—Nano Banana 2 Lite earns its spot in the lineup, and might even be a better option than its more powerful brother.
Daily Debrief NewsletterStart every day with the top news stories right now, plus original features, a podcast, videos and more.
In brief Nano Banana 2 Lite (gemini-3.1-flash-lite-image) generates images in four seconds at roughly $0.034 per image. This means it produces results at about half the cost of Nano Banana 2 at the same resolution and 2.7× faster. In head-to-head testing, the Lite model matched or beat Nano Banana 2 on many fields, but when details are important, the more expensive version may be the better option. Google last week launched Nano Banana 2 Lite—officially gemini-3.1-flash-lite-image—as the entry point in its image generation stack, sitting below Nano Banana 2 and well below Nano Banana Pro. It delivers text-to-image outputs in roughly four seconds, 2.7 times faster than Nano Banana 2, and is positioned as the direct replacement for the original Nano Banana (gemini-2.5-flash-image). The explicit pitch: same Google ecosystem, less money, less waiting.
The model is available through Google AI Studio, the Gemini API, and the Enterprise Agent Platform—and it's baked into consumer products including Search, the Gemini app, NotebookLM, and Google Photos. It works alongside Gemini Omni Flash, Google's new video generation model, through the Interactions API, which lets users stack up to three sequential edits within a single session. The Nano Banana family now reads as a clean three-tier structure: Lite for speed and cost, Nano Banana 2 for the quality-speed balance, Nano Banana Pro for complex professional work.
At roughly $0.034 per image at 1K resolution, Nano Banana 2 Lite is about half the price of Nano Banana 2, which runs $0.067 per image at the same resolution. That puts the Lite model in direct competition with Seedream 5.0 Lite, which comes in at $0.031–0.035 per image. Reve 2.0 undercuts both at around $0.0067 per image via API—though it lacks the deployment breadth that comes with Google's infrastructure. Qwen Image Edit is a good, free, open-source option for standard use cases.
So, is the quality drop from Nano Banana 2 concentrated enough to matter for your specific workflow? Is it distributed enough that most people won't notice?
We ran the same prompts through both models across five categories to find out. The answer is less predictable than you'd expect.
Realism
The realism test is where the gap between Nano Banana 2 and its Lite sibling is most visible. Both models received the same technically demanding portrait prompt: a cinematic image of a 32-year-old female architect on a rooftop at sunset, wearing a beige trench coat and round glasses, holding rolled blueprints specifically in her left hand, with a defocused city skyline behind her, golden hour lighting with a soft rim light, shallow depth of field simulating a 50mm lens, a vertical 4:5 aspect ratio, realistic skin texture, and subtle film grain.
The prompt explicitly frames each element as an independent constraint that can fail.
Nano Banana 2 Lite passed the basic test. The subject is correctly dressed and positioned, wears round glasses, holds blueprints, and stands on a rooftop with a blurred city behind her. But it is slightly, just slightly, less realistic in terms of details: The subject only has one hand, which is oversized in comparison to the rest of the body. The rim light is barely perceptible. Skin texture holds up at thumbnail scale but doesn't survive close inspection. The image, in the end, looks like a competent stock photo, not a cinematic portrait.
Nano Banana 2 produced something photographically different in kind. The subject stands against a fully realized New York City skyline at magic hour, bokeh city lights blooming across the background, a hint of a river visible in the distance. The depth of field is dramatic. The warm rim light clearly separates the subject from the background. The blueprints are in her left hand, not her right hand, as requested.
Both models struggle with symmetry. For example the holes for the buttons and some straps are not consistent, but again, those are details that are spotted upon closer inspection.
For social media content or rapid visual mockups, the Lite version is workable—it communicates the concept. For anything where the image is the final product—a hero image, a client deliverable, a portfolio piece—it will show its seams at any resolution above a thumbnail. Photographic quality is where the Lite model's architecture makes its largest single concession, and it makes it consistently.
Prompt Adherence
Prompt adherence testing used a different strategy: a dense, multi-element scene where each labeled detail functions as an independent failure point. The prompt described a steampunk cityscape viewed from a gargoyle's perch—complete with a hot air balloon labeled "Atlas & Sons Cartographers, Est. 1842," a cable car with a specific named route, a gear-driven clock tower, a gargoyle holding a document labeled "Sector 7 – Condemned," a foreground newspaper with a specific headline, and a detailed Victorian street scene below.
The logic: If a model can hold 10 specific simultaneous constraints, you can trust it on complex creative briefs.
Both models produced visually compelling steampunk scenes. Both correctly place the gargoyle in the foreground, the clock tower at center, the balloon in the sky, and a cable car crossing the frame. At a glance, the differences feel cosmetic—the Lite version is darker and moodier, the full model cleaner and brighter. But the specifics tell a different story. In the Lite version, the balloon reads "Est. 1942" instead of 1842—mostly due to AI grappling to properly render text. The cable car route label is partially garbled. The foreground newspaper headline blurs at the edges, losing legibility on the details that were specifically requested.
Overall, it focused more on visuals than text, which is ok for most use cases.
Nano Banana 2 gets almost everything right. The balloon clearly reads "Atlas & Sons Cartographers Est. 1842." The cable car sign says "Upper Vantis – 4 Stops." The gargoyle holds a document, but the text is illegible. The foreground newspaper reads "Clocktower Falls Silent – City Mourns" in clean, readable type. Every named element appears where it should, with the correct label, in legible form. The compositional decision to use brighter, more editorial lighting also pays off here—it keeps the labeled details readable rather than swallowed by atmosphere.
Casual prompt users won't catch a one-digit transposition on a fictional establishment date. But concept artists, worldbuilders, and narrative illustrators—the people using these models to communicate specific creative logic to clients or collaborators—will notice immediately.
The Lite model's tendency to blur or transpose specific in-image text labels isn't a catastrophic failure, but it introduces a manual correction step that compounds badly at scale.
Spatial Awareness
Spatial awareness testing evaluated how each model handles multi-depth scene composition: multiple objects at close range, a human subject in the middle distance, and atmospheric elements receding into background darkness.
The scene—a medieval alchemist at a cluttered wooden desk, surrounded by an armillary sphere, a lit candle, an hourglass, a skull, star charts, and a glowing green jar, with a black cat silhouetted in an arched window behind him—requires convincing three-dimensional layering to read as coherent rather than assembled.
Both models understood the basic spatial grammar of the scene. Foreground objects are rendered at appropriate scale and shadow detail, the scholar occupies the mid-ground with correct occlusion relationships to the objects around him, and the arched window with the moonlit night sky creates a convincing sense of recession behind the scene. Neither model misplaces objects, collapses depth planes, or introduces spatial contradictions. The scene architecture—front, middle, back—is correctly established in both outputs.
The differences are subtle and real. Nano Banana 2's version has a richer atmospheric depth gradient: The candlelight fades naturally as it reaches the stone walls, the background haziness reads as genuine atmospheric depth rather than digital softening, and the overall scene has a painterly warmth that suggests volumetric space. The Lite version's depth is structurally correct but slightly compressed—the background reads marginally more like a stage flat than a receding room with actual air in it.
At least in this text, the Nano Banana 2 image feels like the same Nano Banana 2 Lite image with a detailed LoRA (a sort of specialized fine tuning layer) applied during sampling.
This is the smallest gap across all five tests. For storyboards, game asset concepts, and most editorial illustration contexts, both models demonstrate adequate spatial reasoning. The Lite model's slightly flatter depth rendering becomes meaningful only in high-resolution output or detailed compositional analysis—and even then, the gap is arguable.
For this category, the Lite model is a viable substitute in the vast majority of practical workflows.
Text Generation
Text generation is where this review produces its most counterintuitive result.
The test prompt described a gritty nighttime hardware store with dozens of simultaneous text elements at different scales and styles: a hand-painted main sign with the store name, founding date, and product categories; a graffiti tag on the façade; window decals with hours and services; a concert poster with band name, venue, date, doors time, and specific ticket prices; a city council meeting notice; a lost cat notice with a phone number; political stickers on a phone booth; and a street parking restriction on the curb.
Text generation at this complexity is difficult because each element has to be correctly rendered while the overall image still reads as a coherent photograph.
Nano Banana 2 Lite actually delivered something genuinely impressive for how fast it is. "KELLERMAN'S HARDWARE & SUPPLY CO. – SINCE 1931 – TOOLS, ROPE, PAINT," graffiti reading "STILL HERE," window signs for "OPEN 7 DAYS / WE BUY SCRAP – ASK FOR RAY / CLOSED," a concert poster for "THE DREDGE PALE MOUTH / SUNDAY JUNE 4 / DOORS 9PM / THE ANCHOR CLUB / $12 ADV – $15 DOOR," stickers reading "THIS MACHINE KILLS FASCISTS" and "JESUS SAVES," a lost cat notice with a specific and legible phone number—every single text element in the prompt is correctly rendered and readable simultaneously in one image.
If there’s something to note, it’s that the image is less realistic. Some posters seem rendered by an editor with poor photoshop skills rather than genuine elements of the scene. One example could be the posters pasted on the phone booth. To be more realistic they should have some natural imperfections, and even deterioration signs. That said, this is a legitimately strong result for any image model, let alone the cheaper, faster one.
Nano Banana 2's version is also strong. Most text is correctly placed and legible, and the overall image reads as a convincing nighttime scene. But the full model's darker, moodier atmospheric rendering—generally one of its assets—works against it here. Several smaller sticker texts fall into shadow and lose legibility. The Lite model's brighter, more neutral lighting, a quality that reads as a weakness in portrait work, becomes a clear advantage when the evaluation criterion is whether all the text in the scene is actually readable.
For text-heavy generation—signage mockups, editorial graphics, product concepts with labeled elements, infographic-style composed images—Nano Banana 2 Lite performs below Nano Banana 2. The model seems to either focus too much on visuals that text becomes garble, or focus so much on text that its placement in scene becomes unrealistic.
ConclusionsNano Banana 2 Lite is not a straight downgrade from Nano Banana 2. It's a focused tool with a specific ceiling, and that ceiling drops hardest in exactly the scenarios where photographic quality is the deliverable, and holds surprisingly steady everywhere else.
Cinematic portrait work, sophisticated lighting physics, fine material texture, close-inspection-quality skin rendering—all of these expose a clear difference between the two models. Style transfer also takes a meaningful hit, not in rendering quality but in contextual comprehension: the Lite model can execute a subject, but it struggles to capture the visual environment in which that subject lives. Prompt adherence degrades specifically on in-image labeled text accuracy—a narrow failure mode, but one that matters badly in worldbuilding, concept art, and any pipeline where specific in-image language carries meaning.
What holds up well—and in some cases holds up better—is specificity: if you require a lot of focus on something, it will make sure everything is there.
Spatial scene architecture, and basic compositional competence are also good. The text generation result warrants specific emphasis: If your workflow involves signage mockups, branded graphics, editorial composites with text-heavy elements, or any pipeline where multiple readable text strings need to coexist in a single image, the Lite model is worth reaching for first. Its brighter rendering defaults, a liability in portrait work, are an advantage when legibility is the metric. Spatially, it handles multi-depth scenes adequately for the vast majority of professional contexts.
On the cost math: at $0.034 per image, Nano Banana 2 Lite runs at roughly half the cost of Nano Banana 2 at 1K resolution ($0.067) and trades almost blow-for-blow with Seedream 5.0 Lite ($0.031–0.035). Reve 2.0 undercuts both dramatically at approximately $0.0067 per image via API, but doesn’t offer the deployment footprint that comes with the Nano Banana ecosystem: Search, NotebookLM, Google Photos, and the Gemini app running off the same model simultaneously.
For teams already inside Google's infrastructure, that integration removes a platform-switching cost that pure-API alternatives can't account for. If you know which use cases you're in—and you're not in the photographic quality bucket—Nano Banana 2 Lite earns its spot in the lineup, and might even be a better option than its more powerful brother.
Daily Debrief NewsletterStart every day with the top news stories right now, plus original features, a podcast, videos and more.
Nano uses a block-lattice architecture where each account has its own blockchain, enabling asynchronous transaction processing that confirms transfers in under one second with zero fees. The network’s fixed supply of 133,248,297 XNO coins was fully distributed via a faucet at launch, with no mining, staking rewards, or inflation, making it entirely deflationary. A February 2026 CoinEx analysis ranked Nano fourth among DAG-based cryptocurrencies, labeling its outlook as niche because it lacks DeFi and smart contract capabilities competitors offer. Nano traded near $0.32 with a market capitalization of approximately $42 million as of July 2026, representing a significant decline from its all-time highs despite stable daily usage. Coinbase launched Nano perpetual futures in July 2025, expanding institutional access, but OKX delisted XNO from its spot market in June 2025, highlighting persistent liquidity risks. Nano (XNO) occupies a distinctive position in the cryptocurrency ecosystem. It is one of the few digital currencies designed exclusively for peer-to-peer payments, with zero transaction fees and sub-second confirmation times.
CoinMarketCap data shows XNO trading near $0.32 with a market capitalization of approximately $42 million as of early July 2026, ranking it outside the top 400 cryptocurrencies. The question for investors and developers is whether Nano’s proven technical efficiency can translate into meaningful adoption before competitors with broader ecosystems absorb its use case.
This article examines the technology, market position, adoption barriers, and competitive landscape that will determine Nano’s trajectory.
Block-Lattice Architecture: How Nano’s Technology Works Unlike Bitcoin or Ethereum, which record all transactions on a single shared ledger, Nano provides each account with its own dedicated blockchain. This block-lattice structure allows users to send and receive funds without waiting for the entire network to process a global block, according to CoinGecko. The result is asynchronous updating, meaning transactions confirm independently and simultaneously.
Network consensus uses Open Representative Voting (ORV), where XNO holders delegate voting weight to representatives who validate transactions. The mechanism is similar to proof of stake but carries no inflationary rewards and requires no token lockups, Forbes notes.
CoinMarketCap states the network can scale to 1,000 transactions per second with appropriate hardware, without requiring an energy-intensive mining network.
Colin LeMahieu, who founded the project as RaiBlocks in 2014 and rebranded it to Nano in 2018, previously worked as a software engineer at Qualcomm, AMD, and Dell, according to the Nano Foundation.
Over 86.5% of XNO’s circulating supply is staked with representatives, indicating strong network participation, CoinMarketCap data shows.
Adoption Barriers and Competitive Landscape A February 2026 analysis from CoinEx ranked Nano fourth among top DAG-based cryptocurrencies, praising its instant, fee-free transactions and improved reliability in 2025. However, the report also characterized Nano as a niche asset lacking DeFi and smart contract capabilities, as CoinMarketCap’s latest updates confirmed.
This creates a structural ceiling: while competitors like Kaspa expand into programmable applications, Nano remains focused exclusively on payments. Community members have highlighted that Nano’s daily active user count exceeded Kaspa’s by 32% on March 1, 2026, yet Nano’s market capitalization remained a fraction of Kaspa’s valuation.
This usage-to-valuation disconnect is the central puzzle for Nano investors. The gap between usage metrics and market capitalization suggests that crypto markets currently reward ecosystem breadth over single-purpose efficiency.
Nano’s refusal to add smart contracts is a philosophical choice that preserves protocol simplicity but limits the network effects that drive valuations for multi-purpose platforms. For Nano to close this gap, it would likely need a catalyst outside the technology itself, such as a major merchant integration or inclusion in a regulated financial product.
Exchange Access and Institutional Exposure Institutional access to Nano shifted in two opposing directions during 2025 and 2026. Coinbase launched Nano perpetual futures for U.S. traders in July 2025, providing regulated derivative exposure.
Interactive Brokers added Nano Bitcoin futures via Coinbase Derivatives in February 2026, CoinMarketCap’s price prediction analysis noted. These products lower entry barriers and signal growing mainstream acceptance.
However, OKX delisted Nano from its spot market in June 2025, highlighting the liquidity risks that smaller-cap tokens face on centralized exchanges. XNO’s 24-hour trading volume fluctuated between $300,000 and $1.6 million in early July 2026, according to CoinGecko.
For context, Bitcoin routinely exceeds $20 billion in daily trading volume. The thin liquidity makes Nano vulnerable to sharp price swings from relatively modest trades. Kraken currently hosts the most active XNO trading pair.
Regulatory Implications Nano’s regulatory standing benefits from its straightforward design. Because XNO was distributed for free via a faucet rather than sold through an ICO, it faces a lower probability of being classified as a security under the Howey test.
No SEC enforcement action has targeted Nano. However, the lack of regulatory classification also means institutional allocators may hesitate to act without explicit guidance from the pending U.S. crypto market-structure legislation.
Nano’s Technological Trajectory Nano’s trajectory depends on whether feeless, instant payments remain relevant as Bitcoin’s Lightning Network and stablecoin rails mature. The Nano Foundation has not published specific roadmap milestones for 2026, leaving the project reliant on community-driven development and organic merchant adoption.
Derivatives expansion through Coinbase provides a structural tailwind, but the OKX delisting underscores that exchange support is never guaranteed. Investors should monitor daily active addresses and trading volumes as the most direct indicators of whether Nano’s technological advantages are translating into durable adoption.
FAQs What is Nano (XNO) cryptocurrency?
Nano is a decentralized digital currency that uses a block-lattice architecture to deliver instant, feeless peer-to-peer transactions, with a fixed supply of 133,248,297 coins and no mining.
How does Nano achieve zero transaction fees?
Each account maintains its own blockchain, enabling lightweight transaction validation through Open Representative Voting without requiring miners or validators to be compensated.
What is Nano’s current price and market cap?
Nano traded near $0.32 with a market capitalization of approximately $42 million as of early July 2026, ranking it outside the top 400 cryptocurrencies by market value.
Why was Nano delisted from OKX?
OKX removed Nano from its spot market in June 2025 as part of a broader delisting of eight tokens, citing liquidity and trading volume thresholds that smaller projects often struggle to meet.
Can Nano support smart contracts or DeFi?
No, Nano is designed exclusively as a payment protocol and does not support smart contracts, DeFi applications, or programmable logic, which limits its ecosystem breadth compared to competitors.
Who created Nano?
Colin LeMahieu, a software engineer with experience at Qualcomm, AMD, and Dell, founded the project as RaiBlocks in 2014 and rebranded it to Nano in 2018.
Is Nano a good investment in 2026?
Nano’s investment case depends on individual risk tolerance, as the token’s small market cap, thin liquidity, and niche positioning create both upside potential and significant downside risk.
Google DeepMind just dropped a new family of image generation models with a name that sounds like it was coined during a late-night brainstorming session fueled by actual bananas. The Nano Banana 2 Lite, officially branded as Gemini 3.1 Flash Lite Image, has landed at the fifth spot on text-to-image leaderboards with an Elo score of 1,255 in evaluations from Artificial Analysis.
It also placed ninth in Multi-Image Edit rankings. For a model designed to be the budget-friendly option in the lineup, that’s a surprisingly strong showing.
What the Nano Banana 2 series actually does The Nano Banana 2 family launched around February 26, 2026, and it comes in multiple variants. The Lite version is positioned as the fastest and most affordable option, aimed squarely at developers and businesses running high-volume image generation tasks.
The feature set across the series is genuinely comprehensive. Conversational multi-turn editing lets users refine images through back-and-forth dialogue rather than starting from scratch each time. Variable aspect ratios mean you’re not locked into square outputs. And upscaling goes all the way to 4K resolution, which puts it in the range of production-quality visual content.
Advertisement
Text rendering, historically one of the weakest points for AI image generators, is listed among the supported capabilities. Anyone who has watched an AI model butcher the word “restaurant” on a storefront sign knows why this matters.
Every generated image gets an invisible SynthID watermark baked in. This is Google’s approach to the growing concern around AI-generated content being passed off as authentic photography or artwork.
On the pricing front, the main variant uses a token-based system. Generating a standard 1K output image requires approximately 1,120 tokens.
The competitive landscape in AI image generation The Pro version of Nano Banana 2 has also been appearing in the top five to seven positions on these same benchmarks, suggesting Google has managed to build a lineup where even the economy option punches above its weight.
The series also integrates real-world knowledge through web search functionalities, allowing the model to pull in contextual information from the web to improve how accurately it represents real-world subjects, landmarks, or concepts.
Subject consistency and instruction adherence were explicitly targeted for improvement in this generation.
And before anyone gets confused: no, this has nothing to do with the meme token called Nano-Banana (NANOBANANA) on the Solana blockchain. The naming overlap is purely coincidental.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.
Several altcoins that were off to a good start in January experienced massive fluctuations as they entered February and March and given the recent market crash, all cryptocurrencies incurred huge amounts of losses. However, altcoins like Stellar and Nano are currently aiming at an upward trend while Steem continues the bearish run.
Stellar
Stellar that has been trending upwards since the start of 2020. The price fell down by 47% after the March 12th crash. However, a 25.3% rise was recorded on April 1; the coin has been maintaining the support at $0.03 since then. CMF indicator which rests at 0.10 hints at a continued upward breakout.
Nano, the 53rd coin as per CoinmarketCap has been trending upwards since March 16, and the coin has recorded an increase breaching all major supports. With the recent price increase by 56%, the coin has continued to maintain support at $0.57. Awesome oscillator indicator which rests above the zero level indicates a possible upward breakout. Additionally, Bollinger Bands seem to contract, which signals lower volatility in the coming days.
Steem was off to a good start at the beginning of 2020, however, the market crash on March 12, pushed the price down by 50%. However, on April 3, the coin rose up by 36% only to start another downward run. The descending triangle pattern formed indicates another downward breakout. Additionally, MACD indicator confirms the downward breakout.
TLDR Arizona advanced Senate Bill 1649 to a full House floor vote after clearing the House Rules Committee. The bill would allow the state to create a Digital Assets Strategic Reserve Fund. The proposal permits Arizona to retain seized cryptocurrencies instead of auctioning them. The legislation names XRP, Bitcoin, Monero, NEAR Protocol, and Nano as eligible assets. Lawmakers set criteria to assess adoption levels and transaction activity for reserve assets. Arizona lawmakers advanced Senate Bill 1649 to a full House vote after clearing the House Rules Committee. The proposal would allow Arizona to retain seized digital assets in a state-managed fund. The measure names XRP, Bitcoin, and Monero as eligible assets under defined standards.
Arizona Crypto Reserve Plan Names XRP as Eligible Asset The House Rules Committee approved SB1649 with eight votes in favor. As a result, the bill now heads to the full House for consideration. Lawmakers introduced the measure to create a Digital Assets Strategic Reserve Fund. The proposal allows the state to keep digital assets obtained through forfeiture or surrender. Currently, agencies auction most seized cryptocurrencies.
State Senator Mark Finchem introduced SB1649 earlier this session. The Senate Finance Committee passed the bill with a 4–2–1 vote. Lawmakers set criteria to determine which assets qualify for the reserve. The criteria review adoption rates, annual transaction volume, and ecosystem development. The bill lists XRP, Bitcoin, Monero, NEAR Protocol, and Nano as eligible assets.
The proposal authorizes the State Treasurer to manage the reserve fund. The Treasurer may invest holdings to generate returns for the state. However, the bill requires that investment actions do not increase financial risk. Lawmakers included this provision to guide fund management practices.
If the House approves SB1649, the bill will move to the governor’s desk. The governor may sign the measure into law or veto it. Lawmakers placed the bill on the House calendar following the committee vote.
Bitcoin and Monero Included in Arizona Reserve Framework SB1649 identifies Bitcoin as a primary digital asset for the reserve. Lawmakers also included Monero under the eligibility framework. The bill groups these assets with XRP under a defined fair value threshold. This threshold evaluates economic strength and technical performance.
Under the measure, Arizona may retain cryptocurrencies received through legal processes. Agencies would transfer those assets to the reserve fund instead of auctioning them. The Treasurer would then oversee storage and management of the holdings. Lawmakers structured the bill to formalize how the state handles digital assets.
The legislation forms part of broader digital asset discussions in Arizona. Lawmakers are also considering Senate Bill 1042. That proposal would allow the state to invest up to 10% of public funds in cryptocurrencies. SB1042 remains under review in the state legislature.
At the federal level, digital asset reserves have also entered policy debates. President Donald Trump signed an executive order establishing a Strategic Bitcoin Reserve. The order also created a broader digital asset stockpile framework. Lawmakers referenced these developments during state discussions.
The House will now determine the fate of SB1649 in a floor vote. If members approve the measure, it will proceed to final executive consideration. The legislative process continues as scheduled in the current session.
TLDR: Counterfeit Ledger Nano S Plus devices use ESP32 chips to steal seeds and PINs in plain text format. A fake Ledger Live app passed Mac App Store review and drained over $9.5 million from 50+ victims. The fraud spans five attack vectors including Android, iOS, Windows, macOS, and physical hardware. Ledger’s genuine check feature fails when hardware is compromised at the supply chain source level. Counterfeit Ledger hardware wallets are at the center of a growing threat targeting cryptocurrency users worldwide.
A security researcher has documented a large-scale operation distributing fake Ledger Nano S Plus devices through multiple online marketplaces.
The compromised units appear identical to legitimate products but carry entirely different internal hardware. Seeds, PINs, and wallet data are being sent directly to attacker-controlled servers, draining any wallet initialized on the device.
Fake Hardware Hides Malicious Chips and Firmware The counterfeit devices replace Ledger’s secure element chip with an ESP32 microcontroller. This substitute chip runs modified firmware labeled “Nano S+ V2 1.”
Unlike the genuine secure element, this hardware stores sensitive data in plain text. That data is then transmitted to remote servers controlled by the attackers behind the operation.
Beyond the hardware, the campaign also distributes a fraudulent version of Ledger Live. This fake app is built with React Native and signed using a debug certificate.
It intercepts transactions and sends sensitive user data to multiple command-and-control servers. Users downloading this version have no visible indication that anything is wrong.
The attack spans five separate vectors: compromised hardware, Android APKs, Windows executables, macOS installers, and iOS apps.
A security researcher just documented a large-scale counterfeit Ledger Nano S Plus operation selling compromised devices across multiple online marketplaces.
The fake units look identical to the real thing but contain completely different hardware. Instead of Ledger's secure… pic.twitter.com/6ZfP9pJkUU
— TFTC (@TFTC21) April 16, 2026
The iOS distribution uses Apple’s TestFlight platform to bypass the standard App Store review process. This approach allows the fraudulent software to reach users without triggering typical security checks. Each channel serves as an independent entry point for the same underlying scam.
Ledger’s built-in genuine check feature is designed to verify device authenticity. However, that verification process can be bypassed when the hardware is tampered with at the source.
This makes the point of purchase a critical security variable. Buying from unauthorized sellers removes the only reliable layer of hardware-level verification.
Separate Mac App Store Fraud Drained Over $9.5 Million Separately, on-chain investigator ZachXBT documented another fake Ledger Live app that passed through Apple’s Mac App Store review. That operation alone drained more than $9.5 million from over 50 victims.
Among those affected was musician G. Love, who lost 5.92 BTC after entering his recovery phrase into the fraudulent application. The app presented itself as the legitimate Ledger companion software.
These two operations together show a clear pattern in how attackers are targeting hardware wallet users. Rather than exploiting firmware vulnerabilities, they are intercepting users before they reach a genuine device.
The fraud happens at the distribution level, not the protocol level. This shift makes user behavior and purchase source more important than ever.
Security best practices remain unchanged despite the evolving tactics. Hardware wallets should only be purchased directly from the manufacturer’s official website.
No legitimate wallet software will ever request a 24-word recovery phrase on screen. Any application asking for seed phrase input is running a scam, without exception.
The broader message from both incidents is straightforward. The hardware itself remains secure when obtained through proper channels.
The vulnerability now lives in the supply chain and software distribution ecosystem. Staying safe requires equal attention to both where a device is bought and how companion software is sourced.
