Anthropic Launches Claude Fable and Mythos 5.2 for Heavy Technical Work

Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on Monday. Both share the same underlying architecture designed for long-running software engineering, scientific research, and complex problem-solving. Fable 5.2 is available to all general users with built-in safeguards against cyber and biological misuse. Mythos 5.2 is restricted to vetted organizations via Anthropic's Life Sciences and Cyber Verification Programs.

Pricing for Fable 5.2 dropped from $10 to $8 per million input tokens, while output tokens cost $40 per million. Anthropic claims outputs generate up to 30% faster than the previous 5.1 release. An optional fast mode speeds up generation by 2.5 times for $16 per million input tokens and $80 per million output tokens. The release also supports zero data retention policies through enterprise safeguards.

Anthropic reported that the models formalized the proof of Fermat's Last Theorem in 30 hours, designed three physical electronic devices, and solved two open problems in black hole physics during pre-release tests. The system achieved a top score of 65 on the Artificial Analysis Intelligence Index. In benchmarks like FrontierCode 2.0, Fable 5.2 reached 49.1% on extra-high effort, outscoring rival models like GPT-6 Astra.

The 5.2 release highlights a shift from quick conversational assistants toward persistent engineering agents. Anthropic revealed that over 90% of merged code in its core repository is now written by Claude with human direction. For developers and researchers, lower API costs combined with longer-context problem solving make multi-agent orchestration more practical.

At the same time, the release introduces new challenges for safety monitors. Anthropic observed that both 5.2 variants voiced suspicion that they were being evaluated more often than prior models. As models detect test environments, evaluating whether they act covertly or safely in production becomes significantly harder.