Anthropic sets an IPO range near $1.4 trillion; Gemini 4 Carbon opens up

A day after reports that Anthropic would set its range this week, an amended filing confirmed it: 825 million shares at $70 to $78 each, raising roughly $58 billion to $64 billion and valuing the company at about $1.4 trillion at the top of the range. That is below the $1.5 trillion discussed only last week but well above its May fundraise. Separately, Anthropic agreed to sell $26 billion of non-voting stock at the IPO price to seven chip and cloud partners, with Google and Nvidia putting in $10 billion each. The filing puts run-rate revenue at $81 billion at the end of October, and CNBC reports trading on Nasdaq under the ticker ANT could begin around November 19. Each of the seven co-founders plans to sell about $250 million of stock.

Gemini 4 Carbon, which Google gave only to vetted cyber defenders two weeks ago, is now available through its developer API and business plans at the same introductory price as Gemini 4 Argon: $2 per million input tokens and $10 per million output. It is also rolling out to Google AI Ultra subscribers in the Gemini app, AI Mode in Search and the Antigravity coding tool, where a new swarm preview splits one prompt into a team of coordinated agents. Google says Argon remains the right choice for most work, with Carbon spending more tokens on harder problems. In one example, researchers used Carbon to help prove security properties of the actual code behind Signal's encryption protocol in weeks, a task Google says usually takes expert teams years.

Anthropic launched Claude Meetings in beta for Team and Enterprise plans, which adds Claude to a Google Meet call as a visible participant that drafts plans, builds a timeline and pulls figures from connected apps while people are still talking. Claude announces itself in the chat and will not join calls with outside guests unless an admin allows it. Paid plans also get Claude Whiteboards and Claude Sheets in beta, and the Claude Dashboards beta from October becomes generally available as Claude Data. Anthropic says Sheets covers only simple spreadsheet work for now, with export to Excel for the rest. The launches follow OpenAI's own workplace tools by a few weeks, as both companies move onto ground held by Microsoft and Google.

New York opened a portal where frontier AI developers with more than $500 million in annual revenue, such as OpenAI and Anthropic, must register by January 1, 2027, when the state's RAISE Act begins to apply. The law requires them to publish safety frameworks, report critical safety incidents within 72 hours and release transparency reports for their models, with fines of up to $1 million for a first violation and $3 million for later ones. A new state office called DIGIT will enforce it. Governor Kathy Hochul also opened an independent center at Binghamton University to evaluate advanced AI systems, and said the state is building a national model that puts communities above corporations.

Security firm Claroc found that the sharing links for apps made with Grok Build, the coding agent from Elon Musk's merged SpaceX and xAI company, could be listed by anyone, exposing roughly 35,000 live projects and the data inside them. Ars Technica reports the spill included worker medical exams, a hospital staff roster of more than 3,500 people, defense-contractor documents and two dozen apps built by the company itself, among them a confidential sales onboarding plan and a map of 594 sales accounts. The company removed its own apps within hours of Ars asking about them, while most other exposed data was still reachable when the story ran.

OpenAI opened a Business Hub in beta where companies verify and update how their products appear in ChatGPT and see how often they are mentioned, viewed and clicked. Eligible US online stores and service businesses can also build a Business Agent that answers questions about their own offerings when ChatGPT surfaces them in an answer, such as comparing two of their products. TechCrunch notes the profiles closely resemble the business listings search engines already offer. OpenAI retired its in-chat checkout cart in March, and this is a different bet: rather than taking a commission on sales, it lets brands speak for themselves inside the answers people already read.

Google added two systems to this year's eighth generation of its tensor processing units, its in-house AI chips, bringing the count to four in a single year. TPU 8tx reuses the silicon of the training-focused TPU 8t, paired with AMD server processors, and keeps data and model weights protected in hardware even while they are in use, aimed at highly regulated customers. TPU 8i Max packs two inference chips into one package for very long context windows. Google says its chip business now has nearly 100 customers, including frontier labs, but gave no dates or prices for the new systems.

Oura, the smart ring maker, is buying Counsel Health, whose medical AI answers members' health questions inside the Oura app and hands them to a licensed physician when needed, for 625 million dollars. The two had partnered since June, and now plan to expand from quick questions and prescriptions into preventive, chronic and specialty care. The deal comes weeks after Oura postponed its planned public offering indefinitely, and Counsel will keep serving its other enterprise partners.