Anthropic released Claude Fable 5.2 and Claude Mythos 5.2, two versions of the same underlying artificial intelligence model designed for the hardest technical work. Fable is available to all users and carries safeguards against misuse in cybersecurity and biology, while Mythos is reserved for vetted organizations working in those sensitive fields. Both produce answers up to 30% faster than the previous versions and cost less.
In Anthropic's benchmark tests, the new models beat the company's previous flagship, Opus 5.5, on almost every measure, and outscored rival systems such as OpenAI's GPT-6 Astra and Google's Gemini 4 Carbon on most. On FrontierCode 2.0, which checks whether a model's code changes would be accepted by real open-source projects, Fable 5.2 reached 49.1%, the highest score of any released model tested.
From black holes to operating systems
During prerelease testing, the models designed three working electronic devices from scratch, solved two open problems in the physics of black holes, and produced a formal proof of Fermat's Last Theorem in 30 hours. In one experiment, Fable 5.2 coordinated more than 1,000 AI subagents for eight days to build a prototype operating system that boots on a laptop. Anthropic says Claude now writes over 90% of the code merged into its core codebase, with human guidance.
Anthropic says its safety testing found the risk of serious misalignment remains low, and that the model performed as well as or better than recent models on most alignment measures. The company also added defenses against distillation attacks, in which someone tries to copy a model by collecting its answers.
The release matters because it shows leading AI models moving from benchmark gains to real scientific and engineering output, including verified mathematics, hardware designs and software built by teams of AI agents. It also raises the stakes of an accelerating race: Google made its competing Gemini 4 Carbon model generally available the day after Anthropic's announcement.