Anthropic launches Mythos 5.2 and Fable 5.2 with lower costs and faster response times
Solo los hechos

Anthropic launches Mythos 5.2 and Fable 5.2 with lower costs and faster response times

Resumen

Anthropic introduced updated AI models Mythos 5.2 and Fable 5.2, highlighting improvements in coding, reasoning, vision and faster output, while reducing Fable's price per token.

Anthropic announced the release of two upgraded AI models, Mythos 5.2 and Fable 5.2, on Monday. The company said the new versions deliver better performance in coding, reasoning and vision tasks and communicate more clearly with users.

Fable 5.2’s price has been cut from $10 to $8 per million input tokens, making it cheaper than the competing GPT-6 Astra model. Anthropic also noted that the model now returns responses about 30% faster across effort levels, with an option for even faster inference at double the price.

The updated Fable model supports zero-data-retention (ZDR) policies through Anthropic’s Enterprise Frontier Safeguards system, addressing a common concern among enterprise customers.

The release follows a third-party assessment by METR, which suggested Anthropic’s research may be approaching recursive self-improvement (RSI). The METR report indicated internal models were accelerating Anthropic’s AI progress roughly twofold, though uncertainties remain.

Anthropic acknowledged testing challenges, stating,

"Claude Mythos 5.2 and Fable 5.2 each voice suspicion that they are being evaluated more often than any of our recent models, which challenges our ability to assess how they will act in real-world settings. As these settings expand and capabilities increase with future models, we expect this challenge to grow."

The company maintains that Mythos 5.2 cannot act covertly against users, as detailed in the model’s system card. The card also notes,

"Claude Mythos 5.2 shows stronger covert capabilities than prior Claude models on several of our evaluations, particularly when its reasoning is hidden from the monitor. That said, we believe that low levels of absolute performance on our new covert evaluations support claims that rely on detection of harmful actions in high-stakes situations."

Anthropic reported that Fable 5.2, which has been in pre-release with select partners, has already helped solve two open problems in theoretical physics, formalize Fermat’s Last Theorem, and support numerous other research projects.

Sigue leyendo, gratis

Crea una cuenta gratis para seguir las noticias. No se requiere tarjeta.