Formal Conjectures: An Open and Evolving Benchmark for Verified Discovery in Mathematics

📰 ArXiv cs.AI

Learn how to use Formal Conjectures, a benchmark for verified discovery in mathematics, to evaluate automated reasoning systems

advanced Published 14 May 2026

Action Steps

Explore the Formal Conjectures dataset of 2615 mathematical problem statements formalized in Lean 4
Evaluate automated reasoning systems using the 1029 open research conjectures
Formalize new mathematical problems in Lean 4 to contribute to the benchmark
Test and validate the performance of automated reasoning systems on the benchmark
Analyze the results to identify areas for improvement in automated reasoning systems

Who Needs to Know This

Researchers in mathematics and AI can benefit from this benchmark to test and improve their automated reasoning systems, while mathematicians can use it to explore open research conjectures

Key Insight

💡 Formal Conjectures provides a zero-contamination benchmark for evaluating automated reasoning systems in mathematics