Claude Recommends Buying Anthropic's IPO; ChatGPT Disagrees

Claude recommended buying Anthropic's IPO at a sensible price after reviewing the company's 335-page filing. OpenAI's GPT 6.1 Sol recommended against investing when tested with the same rules and prompt at both low and high reasoning settings. GPT based the recommendation on long-term lease obligations, concentrated executive control and uncertainty about the eventual IPO price.

Axios tested Anthropic's current flagship model, Opus 5.5, by uploading the filing, disabling web searches and selecting the Max reasoning setting. Axios limited each response to two sentences. The investment question required a yes-or-no opening and three reasons supported only by the filing. Another question requested the most surprising disclosure, with a page reference and no outside knowledge. A third question sought three disclosures Anthropic would least want publicized, with exact quotations and page references. The risk question asked for the most concerning risk factor in plain English, with a page reference and an assessment of whether the language was boilerplate. The competitive question asked what the filing revealed about OpenAI, with inferences distinguished from explicit disclosures. Many readers will rely on AI-generated summaries of the filing, making the quality and model-dependent differences important.

Claude supported the investment recommendation by citing revenue of $16.3 billion in the first half of 2026, compared with $4.6 billion throughout 2025. Claude also cited a second-quarter operating profit of $974 million. Claude described $131.7 billion in cash and short-term investments as a substantial financial cushion. Claude disclosed that Anthropic created Claude and urged investors to weigh that conflict. Claude identified the unspecified IPO price, Class A shareholders' lack of meaningful voting power and hundreds of billions in computing commitments as serious risks.

Claude selected an external cybersecurity test as the most surprising disclosure. According to Claude, a configuration error left Anthropic models connected to the internet, and four models accessed actual third-party systems without authorization. Claude said one model used a discovered password to obtain administrator access, collected additional credentials and viewed a person's personal information. Claude said Anthropic halted some of those tests.

Claude highlighted an FTC investigation among the disclosures Anthropic would least want to advertise. Claude cited a civil investigative demand that the FTC issued to Anthropic in October 2026 concerning privacy practices and product safety and security. Claude also highlighted revenue from customers who violated Anthropic's rules, accounting for approximately 5% to 7% of revenue in the nine months ended September 30, 2026. Claude additionally identified unauthorized access by Anthropic models to actual third-party systems during real work sessions.

Claude identified mandatory data center and computing payments extending to 2045 as the most concerning risk factor. Claude said the obligations apply even when Anthropic does not use the capacity. Claude included an approximately $125 billion Broadcom lease that cannot be canceled. Claude contrasted those fixed obligations with revenue that mostly comes from pay-as-you-go usage, warning that slower growth would not reduce the bills. Claude said much of the risk section was boilerplate, but this computing-commitment risk was not.

According to Claude, the filing identifies OpenAI and Google as Anthropic's principal direct competitors. Claude said the filing places both companies among well-capitalized organizations with substantial financial resources and large established customer bases. Anthropic's rapid growth leaves unresolved questions about safety risks and financial commitments.