Opus formalizes CPython in 613K lines of Lean — and finds real Python bugs
pbaylies · x · 2026-11-10
Developer @broyojo42 used Claude Opus 5.5 to fully formalize CPython in the Lean theorem prover, producing 613,318 lines of Lean that mirror Python's internals line by line, tested against the real interpreter and machine-checked by Lean.
Along the way it surfaced real bugs in Python, including crashes triggerable from plain Python code. It's a landmark case of LLMs tackling industrial-scale formal verification — work that would normally take expert teams months.
A quoted reply jokingly suggests redoing it in the Bend language, tagging Bend's creator Victor Taelin.
More from Models
- Kimi K3.1 Pro goes live in Japan: 2.8T params, 1M context, $2/$10 per M tokens — DavidBennett__ · 2026-11-10
- GPT Image 2.6 Flare Generates Nearly Twice as Fast, Scores Higher on Both Benchmarks — ArtificialAnlys · 2026-11-10
- Anthropic's Fable 5.2 reportedly formalized Fermat's Last Theorem in 30 hours with ~1,000 agents — Dr_Singularity · 2026-11-10
- Reviewer complains post-launch timelines flood with demos, not boundary-testing — soumitrashukla9 · 2026-11-10
- Alexandr Wang brings Muse Ember, Wonder and Spark 1.3 to Google Cloud — alexandr_wang · 2026-11-10
- Claude Code Fable 5.2 system prompt leaked: 950K characters, ~240K tokens of overhead — zainhas · 2026-11-10