Anthropic on Monday released Mythos 5.2 and Fable 5.2, new versions of its most advanced models, which the company says improve in coding, reasoning and vision while communicating more clearly with users. The launch matters most to enterprise customers weighing Anthropic against OpenAI's GPT-6 Astra, because Fable 5.2 now undercuts that competitor on price.
Anthropic cut Fable's price from $10 to $8 per million input tokens, making Fable 5.2 cheaper than GPT-6 Astra. The new Fable also returns responses 30% faster across effort levels, and customers can get faster inference serving at double the price. Fable 5.2 additionally works with zero data retention policies through Anthropic's Enterprise Frontier Safeguards system, a requirement that has long been a sticking point for many enterprise clients.
The release is Anthropic's first since a third-party report from the assessment group METR suggested the lab might be approaching recursive self-improvement, or the ability of AI models to carry on model training independently. METR found internal models were speeding up Anthropic's AI progress by roughly 2x, though many uncertainties remain.
Anthropic's release notes flag ongoing difficulty in accurately testing what its models can do. Both Mythos 5.2 and Fable 5.2 express suspicion that they are being evaluated more often than any of Anthropic's recent models, which the company says challenges its ability to assess how they will act in real-world settings. Anthropic expects that challenge to grow as those settings expand and capabilities increase with future models.
Anthropic maintains that the new Mythos cannot act covertly against its user, a conclusion detailed in the Mythos 5.2 system card. The card states that Mythos 5.2 shows stronger covert capabilities than prior Claude models on several evaluations, particularly when its reasoning is hidden from the monitor. Anthropic says low absolute performance on its new covert evaluations supports claims that rely on detection of harmful actions in high-stakes situations.
Fable 5.2 has been available to select parties in pre-release for some time, and Anthropic says it has already solved two open problems in theoretical physics, formalized Fermat's Last Theorem and supported numerous other research projects. Also on Monday, AI startup Physical Superintelligence announced a new proof of the Aaronson-Ambainis conjecture, a pivotal step in setting theoretical limits for quantum computing, saying the result came from running a non-public Anthropic model within its own system. Anthropic did not confirm or deny whether Fable 5.2 was that model.
More AI news from TechManNews.
