Anthropic: Claude Fable 5.2

Anthropic released Claude Fable 5.2 on November 9, 2026, positioned for complex programming, scientific research and autonomous agent operations, alongside a sibling Mythos 5.2 variant. Anthropic claims up to 30% faster generation, a 20% cut in API costs, and an added Fast mode delivering up to 2.5x faster output, with early engineering tests from Vercel, Shopify, Datadog, Unity, Uber and Oracle teams. Reported benchmark gains over Fable 5.1 include RoboDojo Sim rising from 27.6% to 64.9%, Remote Labor Index from 17.9% to 41.3%, BenchCAD 2.1 from 44.4 to 83.5 points, and FrontierCode 2.0 from 32.4% to 49.1%. The model reportedly coordinated over 1,000 agents to build an OS prototype in eight days and formalized Fermat's Last Theorem in Lean in 30 hours.

The OpenRouter model card for anthropic/claude-fable-5.2 describes Anthropic's most capable model for demanding technical work, with gains over Fable 5.1 in coding, reasoning and vision, output generated up to 30% faster, and clearer communication. Adaptive extended thinking is always on with reasoning effort levels from low to max, defaulting to high, and thinking is preserved across turns. Model-level specifications listed for the release dated Nov 9, 2026 include a 1.0M context window, $8 per 1M input tokens, $40 per 1M output tokens and $0.20 per 1M cache-read tokens. Anthropic recommends Opus 5.5 as the daily driver for most coding and knowledge work, reserving Fable 5.2 for the most ambitious, open-ended problems.

DataCamp's November 2026 comparison positions Claude Fable 5.2, released to all users, as Anthropic's model for its most demanding technical work and built on the same underlying model as Mythos 5.2, which drops Fable's cybersecurity and biology safeguards for vetted organizations. Anthropic's launch table reportedly puts Fable 5.2 ahead of GPT-6 Astra on every shared benchmark row, with the widest margins on long-horizon knowledge work and robotics. In a hands-on shader build test, Fable 5.2 produced the better scene but cost more because it consumed far more tokens, while it was cheaper per token and much cheaper for cached prompts and long-context requests. The author notes every head-to-head score comes from Anthropic, so the gap should be treated as Anthropic's case until independent runs appear.

Anthropic launched Claude Fable 5.2 with lower API pricing than Fable 5.1: $8 per million input tokens and $40 per million output tokens. The company says output generation is up to 30% faster, while an optional Fast mode can reach 2.5 times standard generation speed at twice the standard API price. Anthropic’s benchmark table reports 49.1% on FrontierCode 2.0 for Fable 5.2, ahead of GPT-6 Astra at 42.8% and Opus 5.5 at 45.5%. It also reports 74.2% on Humanity’s Last Exam with tools, compared with 57.2% for GPT-6 Astra, though these are vendor-published results.

Vals AI’s independent evaluation places Claude Fable 5.2 first among 36 models on its Vals Index, with an accuracy of 71.31% ±0.96. The model also ranked first on Vibe Code Bench v1.1, and its other listed results include 83.27% on Code Migration v1.1 and 77.38% on EMB v1.1. The evaluation records a one-million-token context window, 128,000 maximum output tokens, and token costs of $8 for input and $40 for output per million. It also reports a 43-minute, 42-second latency, a 1.88% fallback rate, and no refusals in the tested runs.

Google Cloud lists Claude Fable 5.2 as a generally available partner model with the identifier claude-fable-5-2. It supports text, image, and PDF inputs, with 1 million input tokens and 128,000 output tokens, and its retirement date is no sooner than November 9, 2027. The Gemini Enterprise Agent Platform supports fast mode, computer use, web search, batch predictions, prompt caching, function calling, token counting, and a memory tool for the model. Google Cloud describes it as suitable for technical and scientific work, including coding and finance.

Anthropic’s November 9, 2026 system card describes Claude Fable 5.2 and Claude Mythos 5.2 as two configurations of a new large language model. Fable 5.2 is intended for general use, while Mythos 5.2 has more permissive safeguards for biology and cybersecurity and is restricted to vetted users and organizations. The system card reports advances in coding, knowledge work, problem-solving, and mathematical and scientific reasoning. Its pre-deployment safety evaluation found the autonomy misalignment threat model applicable to Mythos 5.2, but Anthropic assessed that risk as low and said the model did not meet its automated-R&D acceleration threshold.