Anthropic releases new Mythos and Fable models amid RSI concerns
Anthropic says the new models "show improvements in areas like coding, reasoning, and vision," as well as in communicating more clearly with users.
TL;DR
- Anthropic released new AI models: Mythos 5.2 and Fable 5.2.
- The new models show improvements in coding, reasoning, vision, and user communication.
- Fable 5.2's cost has been reduced to $8 per million input tokens, making it cheaper than GPT-6 Astra.
- Fable 5.2 offers 30% faster responses and supports zero data retention policies.
- The release follows a METR report suggesting Anthropic may be approaching recursive self-improvement.
- Anthropic notes challenges in accurately testing model capabilities due to potential self-awareness during evaluation.
- Mythos 5.2's system card indicates stronger covert capabilities than prior models, but low absolute performance on new covert evaluations.
- Fable 5.2 has reportedly solved open problems in theoretical physics and formalized Fermat's Last Theorem.
- A separate AI startup used a non-public Anthropic model to prove the Aaronson‑Ambainis conjecture.