Anthropic releases Claude Fable 5.2 and Mythos 5.2 with leading benchmark scores

Anthropic has launched two new AI models, Claude Fable 5.2 and Claude Mythos 5.2. The company says they are better at coding, reasoning, science, cybersecurity, and robotics, with up to 30% faster output and 20% lower cost.

On the FrontierCode 2.0 coding benchmark, Claude Fable 5.2 scored 49.1%, ahead of OpenAI's GPT-6 Astra and Google's Gemini 4 Carbon. On a multi-subject reasoning test, it scored 74.2% versus 57.2%.

Anthropic also showed the model can handle long projects. It coordinated over 1,000 AI agents and built a bootable operating system from scratch in 8 days, with drivers, file storage, and a window system included.

The model also leads on the VIALS-AA science vision test and the AA-3D benchmark, with a 95% confidence range that doesn't overlap with any other model.