Tuesday, 27 October 2026
Anthropic Files Publicly to List on Nasdaq Under the Ticker 'ANT'
Anthropic Files Publicly to List on Nasdaq Under the Ticker ‘ANT’
On Saturday this site wrote about a leaked draft. On Monday Anthropic made it official: the company filed a public Form S-1 with the US Securities and Exchange Commission for an IPO of its Class A shares, to trade on the Nasdaq Global Select Market as “ANT,” with Goldman Sachs and Morgan Stanley handling the prospectus. No share count, no price range, and nothing can be sold until the SEC declares the filing effective. What is new is what the public document says about customers. As of September 30, more than 11,900 of them spent at least $100,000 a year, more than 1,900 at least $1 million, and about 250 at least $10 million. More than half of the $100,000-plus group now uses both the Claude Platform and an app like Claude Code or Cowork, up from about 30% at the end of last year. Net revenue retention — what existing customers spend now against a year ago — sits around 900%. The filing also lists the risks: the cloud companies that resell Claude increasingly see Anthropic as a competitor, and a few large customers carry a lot of the revenue. A company that sells agents now has to sell itself, to people who will read the risk section.
Source: Anthropic, October 26, 2026
OpenAI’s GPT-6.1 Terra Uses Your Computer at a Tenth of GPT-6 Astra’s Token Price
OpenAI released GPT-6.1 Terra on Monday, a smaller model built for computer use — filling in forms, reading technical drawings, editing video, clicking through repetitive desktop work. It is in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users, and in the API as gpt-6.1-terra at $1 per million input tokens and $5 per million output, half the price of GPT-5.6 Terra and GPT-6.1 Sol. OpenAI’s own numbers: on Agents’ Last Exam, a test of how well agents use work software, it performs on par with GPT-6 Astra at a tenth of the price per token, and it finishes tasks more than three times faster than Astra or Sol. On an internal benchmark its error rate at low effort fell from 53.6% to 16.7%. All of those figures are OpenAI’s, with no independent check yet, and OpenAI still points the hardest coding work at Sol or Astra. One mistake in six at the cheap setting. The price fell faster than that.
Jensen Huang Floats a $2,000-a-Year Fee per GPU for Nvidia’s Agent Software
At GTC Berlin last Thursday, Nvidia’s chief executive said out loud where the company wants its next revenue to come from. “Software licensing will become increasingly a major part of our company’s business,” Jensen Huang said, and gave an example: “about $2,000 per GPU per year to run all the software that Nvidia provides that runs agents on these AI factories.” It was a remark in a Q&A, not a price list. Nvidia has not said which products the fee would cover, which GPUs, or when it would start, and its existing AI Enterprise subscription already costs $4,500 per GPU a year, so the two may not be the same thing. The arithmetic outlets ran — five million licensed GPUs, about $10 billion a year — is theirs, not a forecast. The company that sells the shovels would now like rent on the digging.
Source: Notebookcheck, October 25, 2026