The model scored 49.1% on the FrontierCode 2.0 benchmark, which evaluates coding with AI agents, compared with 45.5% for Opus 5.5 and 42.8% for GPT-6 Astra. In software engineering, it achieved 69.9% on FrontierSWE v2, above Fable 5.1 (56.3%) and Opus 5.5 (62.3%). In the OSWorld 2.1 test, which evaluates computer control, it reached 89.3%.
The results also include advances in scientific research and reasoning. On Terminal-Bench-Science 0.1, Fable 5.2 scored 74.6%, while Opus 5.5 recorded 58.7%. The new model also achieved 74.2% on Humanity's Last Exam and 73.4% on Chartography, which evaluates visual information recognition. According to Anthropic, it also performs complex tasks with fewer tokens and generates responses up to 30% faster than Fable 5.1.
Fable 5.2 also demonstrated the ability to execute practical projects, which go beyond answering questions. In tests, the model designed and programmed functional electronic devices, such as a voice assistant and a virtual pet, controlled robotic arms to open bottles and solve puzzles, and helped scientists with complex calculations about black holes.According to Anthropic, the model is suitable for ambitious technical projects, and Claude Opus 5.5 remains recommended for everyday tasks.
The company also introduced the Claude Mythos 5.2, a version based on Fable 5.2, but with fewer usage restrictions. The model was developed for advanced cybersecurity tasks and biomedical research. Access, however, is restricted to organizations approved by Anthropic’s verification programs for these two areas.
Anthropic makes extra Claude Fable 5.2 usage available to subscribers
To mark the launch, Anthropic made additional usage allowances forFable 5.2 available to eligible subscribers. During the week, users will receive one to three extra limit resets, depending on the length of their subscription.
Additionally, the model was included inup to 50% of the weekly limits for Pro plans and standard Team seats.The measure applies to claude.ai and Claude Code and makes it easier to test the tool on demanding tasks.
API access, for developers and businesses, costs US$ 8 per million input tokens and US$ 40 per million output tokens. There is also a fast mode that speeds up generation by up to 2.5 times and charges US$ 16 and US$ 80 per million tokens, respectively. For individual users, access is through Anthropic’s subscription plans, with no per-million-token charge.
If you enjoyed the content, you may also be interested in checking out “8 differences between the controversial Claude Fable 5 and Mythos 5.”
