ProofBench aims to establish a benchmark suite for mathematical proof verification and generation, enabling meaningful comparison across AI systems. We believe that progress for AI in mathematics needs more than raw performance improvements — it requires deep ideas and key insights from experts in the field. Our goal is to help shape that future by making important mathematical insights available in structured, accessible form.
The first ProofBench workshop takes place from September 22 to October 3, 2025 in Freiberg. Confirmed participants as of June 7:
We are planning to advertise one postdoctoral position in October 2025, with an anticipated starting date in January 2026 or shortly thereafter. More details will follow on this page.
Are you interested in collaborating with ProofBench? We welcome academic and industry partnerships, especially for researchers looking to co-develop or evaluate benchmarks with us. If you are planning a grant application or wish to partially transfer funding to ProofBench, get in touch. We have strategic funds available and institutional support to make competitive offers to strong partners.
ProofBench is currently supported by TU Freiberg, University of Hamburg, and University of Nottingham. We are actively expanding this network and are happy to hear from institutions interested in contributing to the foundational phase of ProofBench.