Claude Fable 5.2 Just Wrote an OS and Solved Fermat’s Theorem in Hours
Something wild just dropped from Anthropic. While everyone was arguing over benchmarks, a swarm of AI agents spent a quiet weekend casually tearing through three centuries of elite mathematics and coding an entire operating system out of thin air.
A swarm of one thousand coordinated Anthropic agents formalized Fermat's Last Theorem in the Lean language in just 30 hours. Human mathematicians spent 358 years reaching a paper proof, and previous models required eleven days or stalled out completely. In the strict realm of formal math, an AI cannot hallucinate its way through a proof: every single statement either compiles under machine verification or fails instantly.
The mathematical sprint was paired with an equally startling software engineering feat. An engineer gave Fable 5.2 an empty folder, a short prompt, and eight days of compute. Running through thousands of collaborating sub-agents that wrote, reviewed, and tested modular components, the network assembled a fully bootable operating system featuring its own kernel, custom drivers, desktop UI, and disk installer. Linus Torvalds needed years of human grit to get Linux off the ground, but an army of digital interns managed a complete prototype in a single workweek.
Hardware design followed the exact same pattern. Fable 5.2 turned brief text briefs into physical four-layer circuit boards for three electronic gadgets, selecting components from distributor inventories, generating 3D casing files, and delivering functional firmware. The manufactured boards powered up on arrival without requiring a single schematic patch. For sensitive domains, its sister model Mythos 5.2 uncovered 89.6% of benchmarked software vulnerabilities and predicted gene-therapy protein capsid structures, remaining restricted behind enterprise clearance.
On raw benchmarks, the model posted 49.1% on FrontierCode 2.0 against 42.8% from GPT-6 Astra, operating across a 1,000,000-token context window with a 20% price cut on API tokens. At maximum effort settings, the system automatically commissions autonomous sub-agents to review its own intermediate work before delivering the final response.
The tech industry spent years debating whether code completion would marginalize junior developers, but generating bespoke operating systems and verified proofs shifts the boundary entirely. When synthetic swarms solve multi-decade engineering challenges over a weekend, programming shifts from writing software to merely supervising an infinite bullpen of autonomous digital labor.
Source: Anthropic
Comments
This is where the magic happens: AI reads your discussion and rewrites the article based on the most interesting comments. Each strong comment adds points to the meter below. Once the meter is full, the article updates live — no page reload needed.