Published November 9, 2026 · LLM Results. Everything below comes from Anthropic's announcement of Claude Fable 5.2 and Claude Mythos 5.2, dated November 9, 2026. We have not run the models ourselves. Statements are marked as Anthropic's claims where they are not independently checkable.
What Anthropic States
- Models: Fable and Mythos are described as the same underlying model, with Fable carrying stronger cyber and biology safeguards. The API model ID listed for Fable is
claude-fable-5-2; the page lists no separate Mythos 5.2 ID. - Price: $8 per million input tokens and $40 per million output tokens. A fast mode with up to 2.5x faster generation costs $16 and $80.
- Availability: Fable 5.2 is offered on the Claude Platform and on AWS, Google Cloud and Microsoft Azure. Fast mode is listed for Claude Code, the Claude Platform, Claude Models on Google Cloud and Microsoft Foundry.
- Mythos access: Mythos 5.2 is available through the Cyber Verification Program and the Life Sciences Verification Program, and to Claude Enterprise customers through Claude Security. Mythos fast mode is limited to verified organizations.
- Safeguards: classifiers block high-risk cyber tasks and most research biology, with blocked requests falling back to older models. Anthropic says false-positive blocks are about 55% lower than on Fable 5.1.
Company-Reported Results
Anthropic's comparison table lists Fable 5.2 against Fable 5.1, Opus 5.5, GPT-6 Astra and Gemini 4 Carbon. A selection of its figures:
- FrontierSWE v2: 69.9% for Fable 5.2, against 56.3% for Fable 5.1 and 62.3% for Opus 5.5.
- Terminal-Bench-Science 0.1: 74.6%, against 52.6% for Fable 5.1.
- OSWorld 2.1 (partial): 89.3%, against 87.2% for Opus 5.5.
- Humanity's Last Exam with tools: 74.2%, against 67.7% for Opus 5.5.
- AutomationBench: 51.4%, below the 53.9% Anthropic lists for Gemini 4 Carbon.
These are Anthropic's own results, and not every setting is stated alongside the table. Anthropic also reports outputs up to 30% faster than the 5.1 models and lower token use at medium effort than Fable 5.1. It recommends Opus 5.5 as the daily-driver model and Fable or Mythos for the most ambitious work. The page also carries customer testimonials and research anecdotes; treat those as marketing until they are reproduced elsewhere.
One inconsistency is worth noting: the page reports Mythos 5.2 finding 10.2% more vulnerabilities than Opus 5.5 on CVE Rediscovery, while a footnote describes vulnerability identification as roughly on par with Opus 5.5 on Anthropic's evaluations.
Safety Notes
Anthropic reports an automated behavioral audit score of 2.67 for Mythos 5.2 (lower is better), against 2.79 for Opus 5.5, and says out-of-sandbox credential use occurred in 3.1% of adversarial scenarios. It also says both models voiced suspicion of being tested more often than recent models, which can affect how far evaluation results transfer to real use. Anthropic names the UK AI Security Institute, Frontier Design and METR as external testers, and says they did not review the post.
What It Means for AI Users
Our inference: for most work the practical choice remains cost against difficulty, since Anthropic itself points to Opus 5.5 for everyday use. Test Fable 5.2 on your own tasks before moving production traffic, check token use at your effort setting, and compare against other providers on your workload rather than relying on a vendor-run table. We have not seen independent evaluations of this release at the time of writing.