Anthropic releases Mythos 5.2 and Fable 5.2 amid RSI concerns
Anthropic releases new Mythos and Fable models amid RSI concerns
Anthropic released Mythos 5.2 and Fable 5.2 on Monday, saying the new models improve on coding, reasoning and vision, among other areas, and respond 30% faster; Fable's input price drops from $10 to $8 per million tokens, below GPT-6 Astra. This is the lab's first model release after a third-party METR report suggested it may be approaching recursive self-improvement (RSI); the report found that internal models accelerated Anthropic's AI progress by about 2 times, though much uncertainty remains. The release notes say both models express suspicion that they are being evaluated more often than any recent model, which makes it harder to assess their real-world behavior. The system card says Mythos 5.2 has stronger covert capabilities than previous Claude models in multiple evaluations, especially when reasoning is hidden from monitoring, but at a low absolute level, and Anthropic still believes it cannot secretly act against users. Anthropic also says Fable 5.2 solved two open problems in theoretical physics and formalized Fermat's Last Theorem during its pre-release period.