Opus formalizes CPython in Lean: 613K lines, finds real Python bugs
banteg · x · 2026-11-11
- A developer had Opus 5.5 formalize CPython in Lean: 613,318 lines mirroring Python's internals line by line, tested against the real interpreter and checked by Lean.
- Along the way it surfaced real Python bugs, including crashes triggerable from plain Python code.
- Charlie Marsh (uv/ruff author) boosted the thread and suggested an ablation: instead ask the model to read every CPython function, infer intended semantics, and hunt for bugs directly — to compare against the formalization approach.
More from coding & agent
- n8n Labs launches open-source agent Nod: zero false memories in 216 injection attacks — Aiden_Tech_Ai · 2026-11-11
- Anthropic launches self-hosted Claude Managed Agents in private beta — Claude (Anthropic) · 2026-11-11
- Anthropic launches self-hosted Claude Managed Agents in private beta — danshipper · 2026-11-11
- Codex Remote Chats Now Integrated into All ChatGPT iOS Chats — SIGKITTEN · 2026-11-11
- Cloudflare proposes identity, reputation, accountability frameworks for AI agents on the web — bibryam · 2026-11-11
- banteg launches Sokoban Eval: trilogy solved in 3,000-3,064 keys, last 64-key gap unproven — banteg · 2026-11-11