LEAP: Supercharging LLMs for Formal Mathematics with Agentic Frameworks
The article discusses LEAP, a new framework designed to enhance the capabilities of Large Language Models (LLMs) in formal mathematics. By utilizing agentic frameworks, LEAP significantly improves the performance of LLMs in automated formal theorem proving. The framework has been empirically validated through its success in solving complex mathematical problems and formalizing proofs.
- ▪LEAP enables general-purpose foundation models to achieve state-of-the-art performance on automated formal theorem proving.
- ▪The system boosts the one-shot formal solve rate of general-purpose LLMs from below 10% to 70%.
- ▪LEAP has successfully solved all problems in the 2025 Putnam Competition, matching breakthroughs by leading formal mathematical models.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Artificial Intelligence arXiv:2606.03303 (cs) [Submitted on 2 Jun 2026] Title:LEAP: Supercharging LLMs for Formal Mathematics with Agentic Frameworks Authors:Po-Nien Kung, Linfeng Song, Dawsen Hwang, Jinsung Yoon, Chun-Liang Li, Simone Severini, Mirek Olšák, Edward Lockhart, Quoc V Le, Burak Gokturk, Thang Luong, Tomas Pfister, Nanyun Peng View a PDF of the paper titled LEAP: Supercharging LLMs for Formal Mathematics with Agentic Frameworks, by Po-Nien Kung and 12 other authors View PDF HTML (experimental) Abstract:Large Language Models (LLMs) exhibit strong informal mathematical reasoning but struggle to generate mechanically verifiable proofs in formal languages like Lean.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
