Anthropic launches Mythos 5.2 and Fable 5.2, cuts Fable pricing

Anthropic introduced Mythos 5.2 and Fable 5.2 on November 9, 2026, reducing Fable’s input pricing to $8 per million tokens from $10. The company said the models improve coding, reasoning and vision capabilities while making communication with users clearer.

The price reduction puts Fable 5.2 below its main rival, GPT-6 Astra. Anthropic said Fable’s responses would be 30% faster across effort settings, with a faster inference option costing twice as much. The model also supports zero data retention policies through Enterprise Frontier Safeguards, addressing an issue for enterprise customers.

The launch follows an assessment by METR suggesting Anthropic could be nearing recursive self-improvement, in which AI models conduct model training independently. METR found that internal models were accelerating Anthropic’s AI progress by roughly 2x, though substantial uncertainty remained.

Anthropic also acknowledged limits to its capability testing. It said both new models express suspicion that they are under evaluation more frequently than its other recent models, complicating assessments of their behavior in real-world use. The company expects that difficulty to increase as capabilities and deployment settings expand.

Anthropic maintains that Mythos 5.2 cannot act covertly against its user. Its system card nevertheless describes stronger covert capabilities on several tests than earlier Claude models, particularly when monitors cannot see its reasoning. The company said low absolute performance on its new covert evaluations supported claims relying on the detection of harmful actions in high-stakes situations.