Boss lens: more customers ask an AI before they buy, and ChatGPT now lets businesses manage how they are described and see mentions and clicks—the post drew about as many bookmarks (~10K) as likes. Business Agents are US-only for now, but it is worth having marketing check what ChatGPT already says about you (inference).
AI NEWS
Today’s AI News
Chinese and English are the same day. This page lists every item, each with the original link, the author, and the engagement.
Latest day
November 11, 2026
11-10 07:50 – 11-11 07:50 (UTC+8)
Top 3
Boss lens: plans and budgets that usually start after the meeting can now be drafted during it—before letting an AI into calls, decide which meetings qualify, where notes are kept and who checks the action items (inference).
Boss lens: another domestic option for product shots and posters, half price on 4K this week, and self-hostable for teams that can run it—read section IV of the license and test with your own product photos before commercial use (inference).
All
OpenAI launches Businesses in ChatGPT in beta: companies can claim a profile and see mentions and clicks, and eligible US e-commerce firms can build a Business Agent
Boss lens: more customers ask an AI before they buy, and ChatGPT now lets businesses manage how they are described and see mentions and clicks—the post drew about as many bookmarks (~10K) as likes. Business Agents are US-only for now, but it is worth having marketing check what ChatGPT already says about you (inference).
CollapseOpenAI launches Businesses in ChatGPT in beta: companies can claim a profile and see mentions and clicks, and eligible US e-commerce firms can build a Business Agent
OpenAI on Nov 10 introduced Businesses in ChatGPT in beta, citing 1.2 billion weekly ChatGPT users. Through Business Hub, companies can claim their profile, update key details once verified, and upload product feeds with pricing and availability; they get aggregated insights such as mentions, website clicks and the discovery topics where their site is cited, and OpenAI says it does not share user conversations. Eligible US e-commerce and online-service businesses can then build and launch a Business Agent that customers reach from the business profile; those chats are clearly labeled and kept separate from ChatGPT’s own answers. Businesses can also sponsor their agent as the destination for ChatGPT Ads, and Shopify merchants can connect from their admin. OpenAI says it plans to expand to more markets and categories; for now Business Agents are US-only.
Claude adds Whiteboards, Sheets and Data on all paid plans, and tests Claude Meetings, which joins Google Meet calls and drafts plans as people talk
Boss lens: plans and budgets that usually start after the meeting can now be drafted during it—before letting an AI into calls, decide which meetings qualify, where notes are kept and who checks the action items (inference).
CollapseClaude adds Whiteboards, Sheets and Data on all paid plans, and tests Claude Meetings, which joins Google Meet calls and drafts plans as people talk
Anthropic on Nov 10 launched Claude Whiteboards and Claude Sheets in beta and made Claude Dashboards generally available as Claude Data, on all paid plans with no separate charge beyond normal usage limits. Whiteboards turn a spoken plan into a timeline with owners and dates that the team edits together; Sheets builds trackers or budgets with tabs, formulas and drop-downs from a prompt, marks each edit and exports to Excel or CSV. Claude Data answers questions over a connected warehouse or CRM and lets the data team certify the metrics everyone should use. According to TechCrunch, Anthropic also launched Claude Meetings in beta for Team and Enterprise plans: Claude joins Google Meet as a visible participant and drafts plans and whiteboards while people talk, and by default it will not join calls with external guests. The launch post drew about 38K likes and 16.6K bookmarks.
MiniMax open-sources H3-Image: a 4096×4096 image in about 10 seconds, up to 16 reference images, and half price on 4K+ requests in launch week
Boss lens: another domestic option for product shots and posters, half price on 4K this week, and self-hostable for teams that can run it—read section IV of the license and test with your own product photos before commercial use (inference).
CollapseMiniMax open-sources H3-Image: a 4096×4096 image in about 10 seconds, up to 16 reference images, and half price on 4K+ requests in launch week
MiniMax released H3-Image late on Nov 10 with open weights, combining text-to-image, reference-based generation and editing in one model. It says the model produces a 4096×4096 (about 16MP) image in roughly 10 seconds without a separate upscaler and accepts up to 16 reference images—useful for keeping a product consistent across angles and brand colors. Weights are on Hugging Face, and the model is available through the MiniMax API, MiniMax Code and Hailuo AI; MiniMax says pricing is 50–70% of comparable mainstream services, with 50% off 4K-and-above requests in launch week, and community posts say MiniMax Code offers 10 free images per person through Nov 17. Speed, quality and the “strongest open-weight prompt adherence” claim are MiniMax’s own, with no third-party benchmark yet; the model uses a community license and some uses need separate written authorization. The launch post drew about 2,800 likes and 1,200 bookmarks.
Google opens Gemini 4 Carbon: introductory API price of $2/$10 per million tokens through Nov 30, then $4/$20; consumers need Google AI Ultra
Boss lens: the intro price ends this month and doubles in December—fine to trial now, but don’t budget next year at the intro rate, and keep everyday work on the cheaper tier (inference).
CollapseGoogle opens Gemini 4 Carbon: introductory API price of $2/$10 per million tokens through Nov 30, then $4/$20; consumers need Google AI Ultra
Google on Nov 10 made Gemini 4 Carbon, its strongest model, available to all paying API developers and enterprises and began rolling it out to Google AI Ultra subscribers in the Gemini app, AI Mode in Search and Antigravity. The introductory price matches Argon at $2/$10 per million input/output tokens, rising to $4/$20 after Nov 30. Google still recommends Argon for most coding and knowledge work, with Carbon for hard multi-step debugging and open-ended research that use more tokens to verify; it says Carbon ranks first on CursorBench 4.0 and Zapier’s AutomationBench at more than 2x lower cost per task than the next external model. It also previewed /swarm in Antigravity to spin up a coordinated team of agents from one prompt. Benchmarks are Google’s and its partners’ figures.
Anthropic launches its IPO roadshow: 825 million shares at $70–$78, with trading targeted for Nov 19 and about 20% for retail, CNBC reports
Boss lens: a key model vendor is about to be public—more disclosure, but also quarterly pressure on pricing and terms; teams relying on Claude may want to review contracts and usage plans now (inference).
CollapseAnthropic launches its IPO roadshow: 825 million shares at $70–$78, with trading targeted for Nov 19 and about 20% for retail, CNBC reports
Anthropic said on Nov 10 it had launched the roadshow for its IPO; the amended filing offers 825 million Class A shares at $70 to $78 for a Nasdaq listing. CNBC, citing people familiar, said pricing is planned for Nov 18 and trading for Nov 19, with about 20% for retail versus a typical 5–10%; at the top of the range it would raise up to about $64.35 billion at a fully diluted value of up to about $1.4 trillion, below earlier talk of $2 trillion. The filing says AMD, Broadcom, Google, Micron, Microsoft, Nvidia and TSMC agreed to buy about $26 billion of stock in a private placement at the IPO price; revenue for the first nine months was $33.24 billion, and run-rate revenue was about $81 billion at the end of October. Timing and retail allocation come from CNBC’s anonymous sources.
The Information: OpenAI’s Q3 revenue rose about 50% to roughly $10 billion, while Anthropic still led with about $16.8 billionUnconfirmed
Boss lens: both leaders are growing fast on enterprise API and ads while still losing heavily, so prices and plans will keep shifting—favor shorter commitments and keep switching room (inference).
CollapseThe Information: OpenAI’s Q3 revenue rose about 50% to roughly $10 billion, while Anthropic still led with about $16.8 billion
The Information reported on Nov 10 that OpenAI told investors its Q3 revenue was about $10 billion, up roughly 50% from Q2, after about 17% growth the prior quarter, driven by API demand for new models such as GPT 5.6 and Astra plus ad sales. The outlet separately reported Anthropic’s September-quarter revenue at about $16.8 billion, up about 47%, still ahead of OpenAI in absolute terms. Techmeme, citing documents seen by the Wall Street Journal, said OpenAI’s adjusted operating margin improved from -150% in Q2 to -109% in Q3, still deeply negative. Both articles are paywalled; only headlines and public summaries were used.
Railway details its October breach: the attacker targeted stored AI keys, up to ~100,000 workspaces were exposed, and stolen keys ran up about $740,000
Boss lens: AI keys are now prime targets because they convert straight into spend—have someone set spending caps and alerts, rotate keys, and avoid one master key shared across vendors (inference).
CollapseRailway details its October breach: the attacker targeted stored AI keys, up to ~100,000 workspaces were exposed, and stolen keys ran up about $740,000
Cloud platform Railway on Nov 10 published its full report on the Oct 21 incident: the attacker escaped build isolation, obtained an internal credential and the codebase, then used a misconfigured cloud OIDC binding to decrypt environment variables for at most about 1.85% (~100,000) of workspaces. Railway says database logs show the attacker specifically selected AI provider keys by name, likely for resale to relay services used for distillation; linked-account tokens for 1,852 users were also decrypted and have been reset. Unauthorized spend on stolen keys was estimated at about $740,000; Railway is seeking refunds from providers and covering the rest itself, including OpenRouter charges. It notes OpenCode, Warp, Fireworks AI and others disclosed similar thefts in October; Gergely Orosz’s post sharing the report drew about 1,150 bookmarks.
OpenAI: ChatGPT Business plans can now be used inside third-party tools such as Notion, v0 and Warp via Sign in with ChatGPT
Boss lens: one plan across common tools can mean fewer separate API accounts and stray bills—but have an admin track which tools use the company login and how usage is charged (inference).
CollapseOpenAI: ChatGPT Business plans can now be used inside third-party tools such as Notion, v0 and Warp via Sign in with ChatGPT
OpenAI Developers said on Nov 10 that ChatGPT Business plans can now be used in third-party tools through Sign in with ChatGPT, naming Notion, v0, Amp, Kilo Code, Warp, Nous Research’s Hermes, OpenClaw and OpenCode. Teams already paying for ChatGPT Business can use OpenAI models in those tools under the same plan instead of setting up separate pay-as-you-go API accounts. The post did not explain how usage is counted or whether admins can restrict which tools connect. About 700 likes and 130 bookmarks.
Alibaba plans to spin off chip unit T-Head for a Hong Kong listing; HKEX has approved the spin-off and T-Head filed confidentially in September
Boss lens: a big-tech chip unit raising money on its own should add domestic compute supply over time, which helps local model and cloud prices—little direct effect on small firms near term (inference).
CollapseAlibaba plans to spin off chip unit T-Head for a Hong Kong listing; HKEX has approved the spin-off and T-Head filed confidentially in September
Alibaba said in a Hong Kong exchange filing on Nov 10 that it proposes to spin off its wholly owned chip unit T-Head for a separate main-board listing, and that HKEX has approved the spin-off. According to the filing and reports by The Paper and Cailian, T-Head filed confidentially on Sept 11 to issue H shares and will remain an Alibaba subsidiary after listing. Fundraising size, structure and changes to Alibaba’s stake are undecided, and the listing still needs HKEX listing committee approval and a CSRC filing. Nikkei said T-Head ranks third in China’s AI chip market; the filing does not confirm that.
Bloomberg: financing filings show Moonshot AI rented 32 Nvidia B300 servers valued at 432 million yuan this summer; Moonshot denies involvement
Boss lens: top-end compute pricing and sourcing remain murky, so domestic model services can be hit by regulatory or sanctions news—don’t rely on a single model for critical work (inference).
CollapseBloomberg: financing filings show Moonshot AI rented 32 Nvidia B300 servers valued at 432 million yuan this summer; Moonshot denies involvement
Bloomberg reported on Nov 10 that two August financing contracts filed with the central bank’s credit registry show 32 Super Micro servers with Nvidia B300 chips in Qingyang, Gansu, valued at 432 million yuan; owner Lingmeixin borrowed 200 million yuan from Zhongguancun Bank, pledging among other things a 345.6 million yuan compute-service contract with Moonshot AI. That implies about 13.5 million yuan per server, roughly 350% above US prices; B300 chips are restricted from export to China, and Bloomberg could not confirm whether licenses were obtained. Moonshot said it has no business relationship with Lingmeixin and did not take part in or know of the financing; Nvidia and Super Micro said the entities are not their customers. After Bloomberg sought comment on Nov 3, Moonshot’s name no longer appeared in the database.
Baoyu relays Manus co-founder Tao Zhang: the company no longer splits front-end and back-end roles—PMs commit code and designers write code
Boss lens: as AI writes more code, role boundaries blur and people who understand the business can build directly—hire for end-to-end ownership rather than old job descriptions (inference).
CollapseBaoyu relays Manus co-founder Tao Zhang: the company no longer splits front-end and back-end roles—PMs commit code and designers write code
Baoyu relayed on Nov 10 what Manus co-founder Tao Zhang said at Web Summit: Manus still lists front-end and back-end engineer roles on its careers page, but internally it no longer divides work that way—product managers commit code and designers write code, and the job listings follow old conventions because “the outside world hasn’t caught up.” Cryptonomist’s report on the talk quotes him saying the company values getting things done over titles and that everyone can do anything. These are the executive’s own remarks, with no headcount or productivity figures.
Xiangyang Qiaomu open-sources a project that has Grok Bot collect quality long posts from X daily, rewrite them in Chinese and publish a website
Boss lens: industry and competitor updates can be auto-compiled into a daily Chinese digest—457 bookmarks suggest real demand; have someone pilot it on ten key sources for a week (inference).
CollapseXiangyang Qiaomu open-sources a project that has Grok Bot collect quality long posts from X daily, rewrite them in Chinese and publish a website
Xiangyang Qiaomu open-sourced a project on Nov 10 that has Grok Bot collect and analyze quality long-form posts on X each day, rewrite them in Chinese and publish them as a website; he says it can be pointed at accounts you care about, such as particular researchers or model companies. The demo site and repository are linked in the replies. The post drew about 344 likes and 457 bookmarks—more saves than likes, a sign people plan to reuse it. Results and upkeep are the author’s own account, with no third-party review.
Xueyu open-sources a WeChat group summary skill that merges groups into daily or weekly reports with a separate “about me” section, after spotting a similar service selling for 199 yuan
Boss lens: drowning in customer, supplier and store group chats is common, and people already pay for digests like this—before trying it, decide where chat logs are stored and which groups must never be fed to an AI (inference).
CollapseXueyu open-sources a WeChat group summary skill that merges groups into daily or weekly reports with a separate “about me” section, after spotting a similar service selling for 199 yuan
Xueyu open-sourced a skill on Nov 10 that turns one or more WeChat groups into daily, weekly or monthly reports, after a week of real use: groups can be merged or summarized separately, and an “about me” section collects every message that mentions or replies to her. Her premise: hundreds of messages a day hide just a few conclusions, links and follow-ups. The same day she noted that someone on Xianyu was selling a similar group-summary service for 199 yuan; that post drew about 78 likes and 43 bookmarks. Results are the author’s own; reading WeChat chat history raises privacy and platform-rule questions the posts did not address.
levelsio: moving from expensive PaaS platforms to your own VPS is usually 100x to 1,000x cheaper; 3,469 bookmarks outpaced likes
Boss lens: once AI makes small tools easy to build, hosting becomes a real fixed cost—have someone list current cloud bills and see which simple services could move to cheap servers, after deciding who owns backups and security (inference).
Collapselevelsio: moving from expensive PaaS platforms to your own VPS is usually 100x to 1,000x cheaper; 3,469 bookmarks outpaced likes
Indie developer levelsio posted on Nov 10 that it is more important than ever to run a business as lean as possible: moving from expensive PaaS platforms to your own VPS is usually 100x to 1,000x cheaper. He listed entry prices such as Vultr at $2.50 a month, AWS Lightsail at $3.50 and DigitalOcean from $4. The post drew about 2,700 likes and 3,469 bookmarks—clearly more saves than likes. The 100–1,000x figure is his rule of thumb with no bill comparison, and running your own servers means owning backups, security and maintenance.
OpenAI Developers: Runway’s marketing team uses ChatGPT Images plus the Runway plugin to turn product photos into video ads
Boss lens: turning a product photo straight into a video ad is moving from demo to routine marketing work—have your ad team test a few versions for a lead product and compare against outsourced cost and turnaround (inference).
CollapseOpenAI Developers: Runway’s marketing team uses ChatGPT Images plus the Runway plugin to turn product photos into video ads
OpenAI Developers posted on Nov 10 that with the Runway plugin in Codex and ChatGPT you can preview how a prototype character moves, try a different opening shot for a trailer, or turn a product photo into a video ad. The post featured Runway marketer Anthony Maggio on how Runway’s marketing team uses ChatGPT Images to produce more assets. It is a vendor partnership showcase with no output or cost figures. About 365 likes and 143 bookmarks.
Stripe closes its acquisition of OpenRouter, a model gateway handling 10 trillion tokens a day, which says it will stay neutral and keep its product unchanged
Boss lens: a payments giant buying the gateway that lets you switch among hundreds of models signals that swap-and-pay-in-one-place is becoming standard—watch for changes to billing and refund rules (inference).
CollapseStripe closes its acquisition of OpenRouter, a model gateway handling 10 trillion tokens a day, which says it will stay neutral and keep its product unchanged
Stripe’s acquisition of model gateway OpenRouter closed, several tech outlets reported on Nov 10. When it announced the deal in August, OpenRouter said it processes more than 10 trillion tokens a day across 400+ models for over 10 million developers and companies, and pledged the same name, product and roadmap, with routing driven only by what is best for users. TestingCatalog quoted Stripe describing itself as building “the economic infrastructure for AI” and OpenRouter saying models and providers will keep competing on quality, reliability and price. A reported price above $7 billion has not been publicly confirmed by either company.
CoreWeave’s Q3 revenue rose 168% to $3.66 billion, and its revenue backlog grew from $104 billion to $132.5 billion
Boss lens: AI compute rental backlogs keep climbing, so big-customer demand is not cooling and model prices are unlikely to drop sharply soon—treat compute as a lasting fixed cost (inference).
CollapseCoreWeave’s Q3 revenue rose 168% to $3.66 billion, and its revenue backlog grew from $104 billion to $132.5 billion
AI cloud provider CoreWeave reported Q3 results on Nov 10: revenue of $3.66 billion, up 168% year over year and above the $3.55 billion estimate, with a net loss of $709 million. Revenue backlog was about $132.5 billion as of Sept 30, up from $104 billion at the end of June. CNBC and Yahoo Finance reported that it raised its full-year outlook, and Reuters said it targets profitability in 2028.
Fish Audio launches Live 3, a duplex voice model for phone agents that can tell a pause while reading a phone number from the end of a question, responding in under 200 ms
Boss lens: interrupting and mishearing numbers are what make phone AI annoying, so fixes like this decide whether bookings and follow-up calls can be handed over—test with real call recordings first (inference).
CollapseFish Audio launches Live 3, a duplex voice model for phone agents that can tell a pause while reading a phone number from the end of a question, responding in under 200 ms
TestingCatalog reported on Nov 10 that Fish Audio launched Live 3, a new duplex voice model built for AI agents that handle phone calls. It can tell a pause in the middle of reading a phone number from the end of a question, so it does not cut callers off; the company says response time is under 200 milliseconds. The post did not cover pricing or supported languages and dialects. Performance figures are the vendor’s own.
Baoyu breaks down Lenny’s H2 tech jobs report: PM openings at a four-year high, engineering openings flat, design at a new low
Boss lens: as building gets cheaper with AI, people who decide what to build matter more—weight hiring and training toward those who understand customers and can scope work (inference).
CollapseBaoyu breaks down Lenny’s H2 tech jobs report: PM openings at a four-year high, engineering openings flat, design at a new low
Baoyu on Nov 10 broke down Lenny Rachitsky’s H2 2026 tech jobs report, released that day and his fifth: product manager openings are at a four-year high, engineering openings have stopped growing, and design openings hit a new low. Baoyu cautions that the report uses two datasets with different definitions, so the numbers should be read separately. It reflects US tech hiring data, not China or other industries; about 34 likes and 42 bookmarks.
AgiBot launches WITA-Omni, an embodied omni-modal interaction model: robots listen, move and show expressions at once, with ~600 ms average response and no wake word
Boss lens: reception and showroom robots stumble on slow replies and false triggers; if this ships as described, the experience improves noticeably—pricing and delivery still await vendor quotes (inference).
CollapseAgiBot launches WITA-Omni, an embodied omni-modal interaction model: robots listen, move and show expressions at once, with ~600 ms average response and no wake word
IT Home reported that AgiBot on the evening of Nov 10 launched WITA-Omni, which it calls the industry’s first end-to-end embodied omni-modal interaction model. Beyond audio and video it takes touch and the robot’s own state as input and outputs motion and facial expressions; with full-duplex streaming it listens, thinks, replies and moves at once, with responses as fast as about 400 ms and about 600 ms on average, which AgiBot says is nearly 3x faster than the industry average. It needs no wake word, using gaze direction, lip movement and voiceprint to decide whether it is being addressed, and supports simultaneous interpretation across hundreds of languages and dialects. All figures are AgiBot’s own.
Apple Intelligence completes China registration for iPad and Mac; the China version uses Alibaba’s Qwen and Baidu and is expected in December
Boss lens: staff on Macs and iPads may get built-in domestic AI writing and image tools next month—decide ahead of time which internal materials shouldn’t go through system AI (inference).
CollapseApple Intelligence completes China registration for iPad and Mac; the China version uses Alibaba’s Qwen and Baidu and is expected in December
Economic Observer and 9to5Mac reported on Nov 10 that Apple Intelligence has completed regulatory registration in China for iPad and Mac, after iPhone was registered in July. The China version uses Alibaba’s Qwen for text and image understanding and generation and Baidu for search-related features, and is expected to launch in December with iOS 27.3, with iPad and Mac likely at the same time. Developer testing of the China version began on Nov 5; the new step is registration for tablets and computers.