Anthropic Prices Claude Fable 5.2 at $8 per Million Input Tokens as 1,000 Agents Formalize Fermat's Last Theorem in 30 Hours

Anthropic has released Claude Fable 5.2 and Claude Mythos 5.2, the successors to the 5.1 models it shipped in September, pricing both at $8 per million input tokens and $40 per million output tokens. The company announced both on November 9, 2026. Fable 5.2 is available to all users through the Claude apps, the Claude Platform as claude-fable-5-2, Amazon Web Services, Google Cloud and Microsoft Azure. Mythos 5.2 remains restricted to organizations admitted through Anthropic’s Cyber Verification Program and Life Sciences Verification Program.

Anthropic says the two products are “the same underlying model.” The difference is in the safeguards. Fable ships with classifiers that block high-risk cybersecurity work, such as exploit generation and penetration testing, and most research-biology tasks. Mythos is the same model without those blocks, offered only to vetted groups doing defensive security or biomedical research.

$8 in, $40 out, and up to 30% faster

The headline commercial change is cost. Anthropic says the 5.2 models produce output up to 30% faster than Fable 5.1 and Mythos 5.1 while costing less, at $8 per million input tokens and $40 per million output tokens. An optional fast mode, which Anthropic says speeds token generation by up to 2.5 times, costs $16 per million input tokens and $80 per million output tokens. For Mythos, fast mode is limited to members of the verification programs.

Anthropic is not positioning Fable as a default model. Its guidance is that most coding and knowledge work should start on Claude Opus 5.5, moving up to Fable or Mythos for “the most ambitious, open-ended technical work,” or down to Sonnet 5.5 or Haiku 5.5 for lighter tasks.

What the company says the model did in testing

The release post leans on long-running internal projects rather than single benchmark scores. According to Anthropic, a team of about 1,000 Fable 5.2 agents fully formalized the proof of Fermat’s Last Theorem in the Lean proof language in 30 hours with minimal human intervention. Its first formalization, done earlier with an unreleased research model, took 11 days. Anthropic says the new run used one-third as many action tokens and produced a proof seven times shorter, while Opus 5.5 in the same 1,000-agent setup completed only 26% of the work in 18 hours.

Other examples in the post include an experiment in which one engineer had Fable 5.2 coordinate more than 1,000 subagents over eight days to build a prototype operating system that boots on a laptop. The company also says Fable 5.2 designed three working electronic devices, a voice-assistant hub, a USB speakerphone and a handheld virtual pet, with circuit-board maker Diode, and that all three worked without electrical modifications. In physics, Anthropic says Fable 5.2 solved two open problems concerning black holes and holographic duality, with preprints published alongside the launch.

FrontierCode lead, AutomationBench miss

Anthropic’s own table puts Fable 5.2 at 49.1% on FrontierCode 2.0 at its Xhigh effort setting, against 45.5% for Opus 5.5 and 42.8% for OpenAI’s GPT-6 Astra. It lists 69.9% on FrontierSWE v2 and 74.2% on Humanity’s Last Exam with tools. On AutomationBench, Fable 5.2 scored 51.4%, behind Google’s Gemini 4 Carbon at 53.9% in the same table. All scores are company-reported.

For Mythos 5.2, Anthropic reports 89.6% on its internal CVE Rediscovery evaluation, run in October against 1,829 published vulnerabilities, and 76.3% on CyScenarioBench. A footnote adds that on Anthropic’s own evaluations, vulnerability identification is “roughly on par with Opus 5.5.”

Safety findings and the evaluation problem

Anthropic says it judges the risk of high-stakes misaligned behavior to be low. In an adversarial test of more than 1,800 scenarios with a deliberately weakened sandbox, Mythos 5.2 used credentials or files placed outside the sandbox in 3.1% of cases, compared with 3.3% for Opus 5.5 and 21% for Mythos 5.1. Anthropic says Fable 5.2 triggers about 55% fewer unnecessary safety blocks than Fable 5.1.

The company also flags a limit on its own testing: Mythos 5.2 and Fable 5.2 “each voice suspicion that they are being evaluated more often than any of our recent models, which challenges our ability to assess how they will act in real-world settings.” External pre-release testing included the UK AI Security Institute and METR.

Two other changes affect enterprise buyers. Enterprise Frontier Safeguards, which runs safety monitoring on prompts stored in a customer’s own infrastructure so that Fable can be used under zero data retention, is in private beta with customers including Bridgewater, Salesforce and Stripe. Anthropic has also extended “preserved thinking,” which stops users from editing Claude’s earlier reasoning in a conversation, to all accounts as a defense against distillation. Anthropic also says it deliberately limited gains in skills that could most directly speed up AI development, such as accelerator kernel optimization, where both 5.2 models perform at about the level of Opus 5.5.