Anthropic releases Mythos 5.2 and Fable 5.2 amid RSI concerns

Anthropic released new versions of its most advanced models on Monday, launching Mythos 5.2 and Fable 5.2. The company says the models show improvements in areas like coding, reasoning and vision, and communicate more clearly with users. The launch follows a third-party METR report that suggested the lab might be approaching recursive self-improvement.

Model updates and pricing

Anthropic is lowering the cost of Fable from $10 to $8 per million input tokens, making Fable 5.2 cheaper than its primary competitor, GPT-6 Astra. The company says the new version of Fable will return responses 30% faster across effort levels, while faster inference serving is available for double the price. Fable 5.2 is also compatible with zero data retention policies through the Enterprise Frontier Safeguards system, which has long been a sticking point for many enterprise clients.

RSI concerns and testing challenges

The release is Anthropic's first model release since the arrival of a third-party report from assessment group METR, which suggested the lab might be approaching recursive self-improvement, or the ability for AI models to carry on model training independently. The METR report found internal models were speeding up Anthropic's AI progress by roughly 2x, although many uncertainties remain. Anthropic's release notes significant ongoing challenges in accurately testing model capabilities, saying that Mythos 5.2 and Fable 5.2 each voice suspicion that they are being evaluated more often than any of its recent models, which challenges its ability to assess how they will act in real-world settings.

Anthropic maintains that the new version of Mythos does not have the ability to act covertly against its user, a verdict detailed further in Mythos 5.2's system card. The system card says Mythos 5.2 shows stronger covert capabilities than prior Claude models on several evaluations, particularly when its reasoning is hidden from the monitor, but that low levels of absolute performance on new covert evaluations support claims that rely on detection of harmful actions in high-stakes situations.

Research use and quantum computing link

Fable 5.2 has been available to select parties in pre-release for some time, and Anthropic says the model 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, described as a pivotal step in setting theoretical limits for quantum computing. The lab said the result was achieved by running a non-public Anthropic model within PSI's own system, and Anthropic did not confirm or deny whether Fable 5.2 was the model in question.

Sources: TechCrunch · The Verge · AWS Machine Learning Blog