Anthropic reports that about 1,000 Fable 5.2 agents formalized the proof of Fermat's Last Theorem in Lean in 30 hours, and that the model solved two open black hole physics problems. The results are Anthropic's own and come from prerelease testing.
Anthropic says its Fable 5.2 model formalized the proof of Fermat’s Last Theorem in 30 hours, with minimal human help. The company also says the model solved two open problems in black hole physics. Anthropic published both claims on November 9, 2026, alongside the Fable 5.2 and Mythos 5.2 release. They come from prerelease testing and have not been independently checked, as far as we can tell.
What Anthropic reports
According to Anthropic’s announcement, a team of about 1,000 Fable 5.2 agents formalized the theorem in Lean, a programming language that checks every step of a mathematical proof by machine. The agents started from minimal prompts and built their own ways of coordinating. They used no mathematical references, only a list of intermediate theorem statements to measure progress.
Anthropic gives this comparison:
- Earlier attempt. Its first formalization, by an unreleased research model, took 11 days and needed a purpose-built harness and human guidance.
- Fable 5.2. It used about one-third as many action tokens and produced a proof seven times shorter.
- Opus 5.5. With the same 1,000-agent setup, it completed 26% of the formalization in 18 hours and made little progress after that.
Both black hole results concern holographic duality, the idea that some gravitational universes can be described exactly by a quantum theory without gravity. Anthropic says Fable 5.2, working largely on its own, finished a calculation for small universes that was outlined in 2003 but judged too laborious. It reports that black holes form abruptly there, as physicists showed in 1983 for large universes. In the second problem, Anthropic says the model found the exact limit of a technique used for more than 15 years to rule out candidate theories of quantum gravity in a simplified universe. Both results are in preprints that Anthropic links.
What is not established
Anthropic’s page does not name a human reviewer for either physics result. Lean’s checking is the verification mechanism it describes for the proof. The company notes that the comparison used a single run per model, and that progress was scored against intermediate statements, not the full proof. Every figure here is Anthropic’s.
What to do with this
Nothing changes for most readers today. Fable 5.2 costs $8 per million input tokens and $40 per million output tokens, as we covered in our price story. If your team runs long agent jobs, the useful signal is the coordination result: Anthropic says a large agent group finished a multi-day task in 30 hours at lower token cost than its earlier attempt.
A strong claim that still needs outside checks
The formalization is checkable in principle, because Lean either accepts a proof or it does not. The physics claims depend on peer review, and preprints have not had it yet. Treat the Opus 5.5 comparison with care too: one run each is a thin basis for ranking models. We will update this story when independent physicists or mathematicians weigh in.