This was the week the AI industry had to put more of its story in writing. On 26 October Anthropic filed its S-1 registration statement to list on Nasdaq under the ticker “ANT”. The filing shows revenue of $16.29 billion for the six months to 30 June 2026, up from $1.16 billion a year earlier, and a half-year net loss of about $80.7 billion, most of which CNBC attributes to accounting charges on convertible notes rather than cash spent. [1][2]
The money picture was not uniformly bullish. TechCrunch, summarising the Financial Times and the Wall Street Journal, reported that OpenAI lowered the target for its new funding round from $30 billion to $20 billion, with about $11 billion secured. Those are press reports, not company figures. Read together, a growing audited revenue line on one side and a smaller round on the other say more than any valuation headline. [3]
The same week brought two model launches with opposite release strategies. OpenAI made GPT-6.1 Terra available across ChatGPT Work, Codex and its API on 26 October, pitching cheaper and faster “computer use”. Google DeepMind announced Gemini 4 Carbon on 28 October but limited access to vetted cyber defenders, citing its vulnerability-finding ability; it gave no date for wider release. Both companies’ benchmark claims are their own. [4][5][6]
Independent measurement lagged behind the launches. On the maintainer-run Terminal-Bench 4.0 leaderboard, the top verified entry remained Claude Code with Opus 5.5 at 64.8%, while OpenAI’s self-reported 50.3% for Terra has not yet been checked there. Agent research outside the US moved too: DeepSeek’s Harness team released an MIT-licensed “super-agent” prototype, and a Tsinghua–Zhipu-led paper reported cheaper, better scores by letting agents tune their own scaffolding. [7][8][9]
Open weights had a European moment. Mistral published the weights of Mistral Large 4, nicknamed “Le Chonk”, on Hugging Face and pitched it as “built for sovereignty”. Its ranking claims are the company’s own, but the release keeps a capable open model available to governments and firms that do not want to depend on US cloud APIs. [10][11]
For publishers, OpenAI, Google, Microsoft and Anthropic promised on 30 October to honour a single, standard opt-out from AI training, on staggered timelines: Microsoft from 1 February 2027, Google sometime in 2027, OpenAI “in the coming months”, and Anthropic saying it has already started. The signal covers future training only, leaves scraping allowed by default, and may reduce visibility in the companies’ AI products. We have added the dated pledges to our promise tracker. [12]
What to watch next: the price range and valuation in Anthropic’s amended filing; whether Gemini 4 Carbon reaches general availability and on what terms; independent Terminal-Bench results for Terra; and whether the first opt-out deadlines are met.