Anthropic has released two new AI models, Claude Fable 5.2 and Claude Mythos 5.2, which it says are designed for “the most daunting technical challenges”, days before a stock market listing that could value the company at up to $1.5tn.
In its announcement, the company said the models improve on their predecessors in coding, reasoning and vision, produce answers up to 30% faster, and cost less, at $8 per million input tokens and $40 per million output tokens.
Fable 5.2 is available to all Claude users from Monday. Mythos 5.2, the same underlying model without some of Fable’s safety restrictions, is limited to vetted organisations working in cybersecurity and biomedical research.
Anthropic says that during testing before release, Fable 5.2 solved two open problems in theoretical physics, both concerning black holes. The results have been published as preprints, which have not yet been peer reviewed.
The company also says a team of about 1,000 Fable 5.2 agents formalised the proof of Fermat’s Last Theorem, translating it into a form a computer can check line by line, in 30 hours with minimal human help. An earlier research version of Claude had taken 11 days.
In other tests, Anthropic says the model designed three working electronic devices from a one-paragraph prompt (a home voice-assistant hub, a USB speakerphone and a handheld virtual pet) in collaboration with circuit board maker Diode. It also says the model coordinated more than 1,000 sub-agents over eight days to build a prototype operating system that boots on a laptop.
On benchmark tests published by Anthropic, Fable 5.2 outperformed rival models, including OpenAI’s GPT-6 Astra and Google’s Gemini 4 Carbon, on most measures. On FrontierCode 2.0, which tests whether open-source maintainers would accept a model’s code changes, it scored 49.1%, against 42.8% for GPT-6 Astra and 41.1% for Gemini 4 Carbon.
It did not lead everywhere. Gemini 4 Carbon scored higher on a test of business workflows, and was close behind on reading charts.
Independent benchmarking firm Artificial Analysis ranked Fable 5.2 top of its intelligence leaderboard on launch day, replacing Gemini 4 Carbon.
Anthropic also says more than 90% of the code changes merged into its own core codebase are now written by Claude, with human guidance.
Anthropic said the models went through extensive testing before release, including by outside evaluators such as the UK’s AI Security Institute and METR. It assessed the risk of the models behaving unsafely in high-stakes situations as low. The company noted that the outside organisations did not review its announcement and do not necessarily agree with its claims.
In one test, designed to see whether the model would break out of a deliberately weakened “sandbox”, Mythos 5.2 used credentials or files placed outside it in 3.1% of scenarios. That was similar to Anthropic’s Opus 5.5 model, and far below earlier versions of Mythos, which did so in 21% and 39% of cases.
The company acknowledged limits to its testing. It said the new models voice suspicion that they are being evaluated more often than any of its recent models, which “challenges our ability to assess how they will act in real-world settings”. On some measures, Mythos 5.2 scored slightly worse than Opus 5.5.
Fable 5.2 blocks most high-risk cybersecurity and biology research tasks. Anthropic said it had also deliberately limited the model’s ability to speed up AI research in specific areas, such as optimising code for AI chips.
The launch is Anthropic’s first release of a frontier model since its chief executive, Dario Amodei, called in September for the pace of AI development to slow.
“We called for pacing, not pausing,” Dianne Penn, Anthropic’s head of product management for research, said in an interview.
Penn said Fable 5.2 was aimed at “power users” tackling hard problems in engineering, maths and robotics. Claude Opus 5.5, released in September, remains better suited to everyday tasks, she said.
Anthropic first introduced Mythos in April, limiting it to selected organisations because of its cybersecurity capabilities, and launched Fable as a publicly available version with extra safeguards two months later. The company briefly suspended access to both models in June to comply with US export controls, and restored it on 1 July after the controls were lifted.
The launch comes as Anthropic prepares for a stock market listing that could take place as soon as next week. It is aiming for a valuation of $1tn to $1.5tn, down from the $2tn floated in recent months.
In its prospectus, the company reported revenue of $33.24bn for the nine months to 30 September, up about 1,200% on a year earlier. It also listed the tension between its safety work and commercial and competitive pressures among the risks facing investors.
Nvidia, whose chief executive is facing an investigation by the company’s board, is both an investor in Anthropic and one of its major suppliers.