On Nov 10, Anthropic released two frontier AI models, Claude Fable 5.2 and Claude Mythos 5.2. The company says the new models bring significant gains in coding, reasoning, scientific research, cybersecurity and robotics, while delivering up to 30% faster output and 20% lower cost.
On FrontierCode 2.0, a benchmark for agentic coding, Claude Fable 5.2 scored 49.1%, beating OpenAI's GPT-6 Astra (42.8%), Google's Gemini 4 Carbon (41.1%) and Anthropic's own Opus 5.5 (45.5%).
Leads multiple benchmarks, with a clear edge in agentic coding
Beyond FrontierCode 2.0, Claude Fable 5.2 stood out across several benchmarks. On Humanity's Last Exam, a multidisciplinary reasoning test, Fable 5.2 scored 74.2%, ahead of GPT-6 Astra's 57.2%. On VIALS-AA, a scientific visual reasoning benchmark, Claude Fable 5.2 (top configuration, with fallback) scored 67%, above Gemini 4 Carbon (high) at 64% and GPT-6 Astra (max) at 63%.
| Benchmark | Claude Fable 5.2 | GPT-6 Astra | Gemini 4 Carbon | Claude Opus 5.5 |
|---|---|---|---|---|
| FrontierCode 2.0 | 49.1% | 42.8% | 41.1% | 45.5% |
| Humanity's Last Exam | 74.2% | 57.2% | — | — |
| VIALS-AA | 67% | 63% | 64% | — |
On AA-3D, which measures a model's ability to build interactive 3D experiences, Claude Fable 5.2 (Max) earned an Elo score of 1,271, 67 points above Claude Sonnet 5.5 (1,204) and Claude Opus 5.5 (1,202). Its 95% confidence interval (1,246 to 1,297) does not overlap with any other model, putting its lead beyond the margin of error.
Notably, Claude Fable 5.2 made more tool calls on VIALS-AA, in sharp contrast to OpenAI models, which make fewer tool calls as their intelligence rises.
Long-horizon demo: an eight-day build of a bootable operating system
Anthropic demonstrated Fable 5.2's long-horizon collaboration: the model coordinated more than 1,000 AI agents over eight days to build a bootable operating system from scratch, complete with device drivers, file storage, a window system and an installer.
Prompts shape code-structure decisions
According to Simon Willison, citing Anthropic documentation, Claude Fable 5.2 responds markedly to how a prompt frames code ownership. In Anthropic's tests on small coding tasks, when a prompt said nothing about ownership or presented the code base as someone else's, the model tended to add to the existing structure, even when that repeated existing code.
This codebase is yours: what it's for and how it's built are your call, and you'll keep building on it.
Anthropic's tests show that with the sentence above, the model consolidated duplicate code in about half of attempts; without it, the rate was under one in ten.
What to watch
Anthropic has not disclosed release dates or pricing for the two models. The industry is watching how the agentic coding gains will reshape competition among leading models.