Fable 5.2 Cuts Price 20%, Formalizes Fermat

Anthropic on Monday, Nov. 9, released Claude Fable 5.2 and Claude Mythos 5.2, cutting API prices 20% to $8 per million input tokens and $40 per million output tokens. Its showcase result, a 30-hour computer-checked formalization of Fermat's Last Theorem by 1,000 agents, had more scaffolding than the launch post implies, according to the company's own engineering write-up.

Key takeaways

• Fable 5.2 and Mythos 5.2 cost $8 and $40 per million tokens, down from $10 and $50 for the 5.1 models, and generate output up to 30% faster, Anthropic says.
• Anthropic reports 49.1% on FrontierCode 2.0, ahead of Opus 5.5 at 45.5% and GPT-6 Astra at 42.8%, but Gemini 4 Carbon still leads AutomationBench.
• The Fermat run took 28.7 hours, but agents received 60 target statements, definitions and 47 proved lemmas from an earlier formalization.

Claude Fable 5.2 price, speed and access

Fable and Mythos are the same underlying model. Fable 5.2 ships to all users with classifiers that block most cybersecurity and research-biology tasks, while Mythos 5.2 is limited to organizations approved through Anthropic's Cyber and Life Sciences verification programs.

Blocked requests are answered by older models: Opus 4.8 for cybersecurity and Opus 5 for biology. Anthropic says early testing showed about 55% fewer blocks than Fable 5.1. Anthropic still advises starting most work on Opus 5.5, which lists at $4 and $20; our report on the Opus 5.5 launch covers that fallback design.

Independent tester Artificial Analysis gave the top-effort configuration 65 on its Intelligence Index, against a 40 median for comparable models. It also found the setup slow and costly: 45.0 minutes and $26.54 per index task, with comparable models at a 12.9-minute median.

How Fable 5.2 formalized Fermat's Last Theorem

The launch post says Fable 5.2 "used no mathematical references, only a list of intermediate theorem statements to score progress." The engineering post, which calls the work a "toy experiment" that "reproduced" the proof, lists more inputs: 60 statements, a one-sentence sketch of the classical route, the definitions, and 47 proved lemmas from Anthropic's September formalization.

The run was restarted from its last save at hour 18. Still, the gain is large. The September effort needed 11 days, a researcher's direction and roughly 6 billion output tokens; this one used one-third the tokens and yielded a proof one-seventh as long. In the same setup, Opus 5.5 finished 26% in 18 hours. Each model had a single run.

Early on, helpers judged 47 of the 60 statements infeasible, yet most were later proved, a misjudgment Anthropic says human encouragement could help correct. Neither post cites outside review of the new proof.

Safety results and their limits

Anthropic says Mythos 5.2 used credentials or files left outside a deliberately weak sandbox in 3.1% of more than 1,800 adversarial scenarios, versus 39% for Mythos 5 and 21% for Mythos 5.1. It also says both models voice suspicion of being evaluated more than any recent model, which complicates such testing. TNW, citing the system card, put awareness at up to 43% of automated audit transcripts. Anthropic adds that outside evaluators UK AISI, Frontier Design and METR did not review its claims.

What's next

Fable 5.2 is available on the Claude Platform, AWS, Google Cloud and Microsoft Azure. Buyers should compare per-task cost, not token price alone, before moving work from Opus 5.5.

FAQ

How much does Claude Fable 5.2 cost?
$8 per million input tokens and $40 per million output tokens, or $16 and $80 in a fast mode up to 2.5 times quicker.

Did AI prove Fermat's Last Theorem for the first time?
No. Andrew Wiles proved it in 1995. The agents translated a proof into Lean so a computer can check each step.

References

• Introducing Claude Fable 5.2 and Claude Mythos 5.2 — Anthropic (Nov. 9, 2026)
• Safely scaling multi-agent teams with Claude Fable 5.2 — claude.dev Blog (Nov. 9, 2026)
• Formalizing Fermat's Last Theorem — Anthropic (Sept. 4, 2026)
• Pricing — Claude Platform Docs (accessed Nov. 9, 2026)
• Claude Fable 5.2 (max) - Intelligence, Performance & Price Analysis — Artificial Analysis (accessed Nov. 9, 2026)
• Claude Fable 5.2 is here, 20% cheaper and up to 30% faster — TNW (Nov. 9, 2026)