Published November 13, 2026 | Version v1

LPBench: A Lean Benchmark for Research-Level Mathematical Theorem Proving

  • 1. TU Freiberg
  • 2. Universität Hamburg

Description

In October 2026, OpenAI released hundreds of manuscripts resolving open mathematical problems, without publishing the full set of problems attempted by its model. Assessing the broader capabilities and limitations of such systems calls for independent benchmarks.

To this end, we introduce LPBench, a benchmark for research-level mathematics comprising 117 problems formalised in Lean 4, primarily in combinatorics and graph theory. As a first application, we evaluate Claude Opus 5.5 and Claude Fable 5.1, each with and without web search.

Website: https://lp-bench.github.io/

Notes

Supported by the Krüger Stiftung and by Anthropic's AI for Science program. MSC2020: 68V15, 68V20, 05-04, 05C99.

Files

LPBench.pdf

Files (393.8 kB)

Name Size
md5:5e345e63568b7bc961b777b355abeaaa
393.8 kB Preview Download

Additional details

Related works