Anthropic launches Claude Fable 5.2 and says it formalized Fermat's Last Theorem in 30 hours

Anthropic released Claude Fable 5.2 on November 9, 2026, and says a team of 1,000 Fable 5.2 agents formalized the proof of Fermat's Last Theorem in 30 hours. The model is available now at $8 per million input tokens and $40 per million output tokens. Its sibling, Claude Mythos 5.2, goes only to vetted organizations.

What are Claude Fable 5.2 and Mythos 5.2?

Two products built on one model. In Anthropic's words, "Claude Fable and Claude Mythos are the same underlying model." The difference is the safeguards. Fable adds stronger cybersecurity and biology protections, so Anthropic can offer it to everyone. Mythos goes to vetted organizations doing cybersecurity or biomedical research.

Claude Fable 5.2 is available on all platforms, Anthropic says, including Amazon Web Services, Google Cloud and Microsoft Azure. The API model ID is claude-fable-5-2. The Verge confirms Fable 5.2 launched on November 9, while Mythos 5.2 starts with vetted organizations only. CNBC reports that access runs through Anthropic's Life Sciences and Cyber Verification Programs.

There is one side door. Anthropic says Claude Enterprise customers can use Claude Security, which runs on Mythos 5.2, without joining the cyber program.

The 5.2 pair follows Fable 5.1 and Mythos 5.1, which arrived in September, according to The Verge. Anthropic also says Fable 5.2 blocks about 55% fewer requests than Fable 5.1. When it does block a high-risk cyber request, Opus 4.8 answers instead. For blocked biology requests, Opus 5 does.

What does "formalized Fermat's Last Theorem" mean?

It means Claude rewrote an existing proof so that software can check every step. It does not mean Claude found a new proof. Formalization reproduces Andrew Wiles's existing proof in machine-checkable form; it is not new mathematics.

Fermat's Last Theorem says no three positive integers a, b and c satisfy aⁿ + bⁿ = cⁿ for any whole number n above 2. Andrew Wiles published the accepted proof in 1995.

Formalizing it means writing the whole argument in a proof assistant, a program that accepts a proof only if each step follows from the last. Anthropic names the tool: the Lean proof assistant. Once a proof passes the checker, people can trust it without checking every line by hand. Anthropic gives exactly that reason for the work.

So the 30-hour figure measures how fast an AI system turned human mathematics into a machine-checked proof. That is a hard engineering task, but it is a different claim from "AI proved Fermat."

How did 1,000 agents do it in 30 hours?

By Anthropic's account, a "team of 1,000 Fable 5.2 agents fully formalized the theorem in just 30 hours, with minimal human intervention." The agents started from minimal prompts and built their own ways to coordinate. They used no mathematical references, only a list of intermediate theorem statements to track progress.

The benchmark is Anthropic's own earlier attempt. An unreleased research version of Claude did the first formalization, and it took 11 days. That run used a custom-built agent setup and guidance from a human researcher. Compared with it, Fable 5.2 needed about a third as many action tokens and produced a proof seven times shorter, Anthropic says.

Anthropic also ran Opus 5.5 in the same 1,000-agent setup. It completed 26% of the formalization in 18 hours and made little progress after that. The company flags two caveats. Each model got a single run, and both teams resumed from saved states at the 18-hour mark for 12 more hours.

The Verge carries the 30-hour claim as Anthropic's statement, and we found no outside review of the Lean proof in the launch coverage. For a related case, our report on Claude's formal verification of 123 libraries shows how machine-checked proofs work in software.

How fast is Claude Fable 5.2, and what does it cost?

Anthropic says Fable 5.2 and Mythos 5.2 "generate outputs up to 30% faster" than the 5.1 versions, and CNBC reports the same 30% figure. Standard pricing is $8 per million input tokens and $40 per million output tokens, according to both Anthropic and CNBC.

A fast mode costs twice as much: $16 per million input tokens and $80 per million output tokens. In return, Anthropic says it speeds up token generation by up to 2.5 times. Fast mode runs in Claude Code, the Claude Platform, Claude models on Google Cloud and Microsoft Foundry. For Mythos, fast mode is limited to organizations in the verification programs.

Anthropic calls 5.2 "available at a lower price" but gives no number against 5.1 on its launch page. Subscribers get something too. Anthropic says eligible paid plans get extra Fable usage this week. On Pro plans and standard Team seats, Fable can take up to 50% of weekly usage limits.

How does Claude Fable 5.2 compare with Opus 5.5, GPT-6 and Gemini?

On Anthropic's own benchmarks, Fable 5.2 leads on most tests but not all. These are vendor numbers, run at the highest available effort unless noted, so treat them as a claim to test.

Gemini 4 Carbon still beats Fable 5.2 on AutomationBench, 53.9% to 51.4%. Cost per task matters as much as raw scores, though. On FrontierCode 2.0, an agentic coding test, Anthropic says Fable 5.2 at High effort scores about level with Opus 5.5 at Max effort, for roughly half the cost per task.

C

Claude

49

Assistant focused on long documents and code work

🇺🇸United States·Added 2mo ago

Freemium·from $17/mo·API

88trust score out of 100

C

Claude Code

70

Anthropic's coding agent that reads a repository, edits files and runs commands from the terminal, IDE, desktop app, browser or Slack

🇺🇸United States·Added 2mo ago

Paid·from $17/mo·API

96trust score out of 100

1

C

ChatGPT

70

General assistant with tools, memory, files and voice

🇺🇸United States·Added 2mo ago

Freemium·from $8/mo·API

89trust score out of 100

2

G

Gemini

71

Google assistant wired into Workspace and Android

🇺🇸United States·Added 2mo ago

Freemium·from $4.99/mo·API

94trust score out of 100

If you are weighing chat assistants rather than raw models, our Claude and Gemini comparison covers plans and features.

Why does the timing matter?

Because of what Anthropic's CEO said two months ago. CNBC reports this is the first time Anthropic has released frontier models since Dario Amodei called for slowing the pace of AI development in September.

Anthropic's framing is that the call was about speed, not a stop. "We called for pacing, not pausing," Dianne Penn, Anthropic's head of product management for research, told CNBC. The safeguards split between Fable and Mythos is part of that pitch. The most capable cyber and biology uses stay behind vetting, while everyone else gets the filtered model.

What we make of it

We read Anthropic's post next to the CNBC and Verge coverage, and each core number has a second source: the November 9 launch, the $8 and $40 prices, the 30% speed figure and the 30-hour run. The Fermat details beyond that, from the 1,000 agents to the 11-day first attempt, come from Anthropic alone. For day-to-day work, the more practical news is the lower cost per task. If you pick AI coding assistants by price and speed, Fable 5.2 is worth a test run on your own tasks before you trust the charts.

What to watch next

Three things. First, whether Anthropic publishes the Lean proof so outside mathematicians can inspect it. Second, independent benchmark runs of Claude Fable 5.2 against GPT-6 and Gemini 4 Carbon. Third, how fast Anthropic widens Mythos 5.2 beyond its verification programs.