Tag
A new Google paper introduces LEAP, an agentic framework that enables general LLMs to solve formal math problems by planning proofs and checking each step, raising performance from under 10% to 70% on the Lean IMO benchmark and solving all 2025 Putnam problems.