Data Party THU

Jul 3, 2026 · Artificial Intelligence

From <10% to 70%: How Google’s Iterative Proof Framework Cracked the Putnam Competition

Google’s LEAP framework transforms generic LLMs into an agentic system that iteratively builds proof blueprints, boosting formal theorem‑proving success on Lean‑IMO‑Bench from under 10% to 70% and achieving a perfect 12‑out‑of‑12 score in the 2025 Putnam Competition.