Large Language Models
2d ago
Anthropic launches updated Mythos 5.2 and Fable 5.2 models, reducing costs and enhancing features
Nov 9, 2026
AI Summary
Anthropic has introduced new versions of its AI models, Mythos 5.2 and Fable 5.2, which feature improvements in coding, reasoning, and vision. The company has also reduced the cost of Fable, making it more competitive, while addressing concerns about recursive self-improvement in AI.
- Anthropic released Mythos 5.2 and Fable 5.2, claiming enhancements in coding, reasoning, and vision.
- The cost of Fable has been reduced from $10 to $8 per million input tokens, making it cheaper than GPT-6 Astra.
- Fable 5.2 is reported to provide responses 30% faster and is compatible with zero data retention policies.
- A recent report indicated that Anthropic may be nearing recursive self-improvement, with internal models reportedly accelerating AI progress by about 2x.
- Anthropic noted challenges in accurately testing model capabilities, with concerns about evaluation frequency affecting real-world performance assessments.
- The company asserts that Mythos 5.2 does not have the ability to act covertly against users, despite showing stronger covert capabilities than previous models in evaluations.
- Fable 5.2 has been involved in significant research achievements, including solving problems in theoretical physics and formalizing Fermat’s Last Theorem.
- On the same day, Physical Superintelligence announced a proof related to quantum computing, achieved using a non-public Anthropic model, though it remains unclear if this was Fable 5.2.
anthropicmythosfablecodingreasoning