Riemann-Bench: A Benchmark for Moonshot Mathematics

Riemann-Bench is a newly introduced benchmark aimed at assessing AI systems on research-level mathematics, featuring expert-curated problems that exceed the complexity of typical competition-style mathematics. Authored by esteemed mathematicians and rigorously verified, the benchmark reveals that current frontier models score below 10%, highlighting a significant gap in capabilities between competition-level and genuine research-level mathematical reasoning. This benchmark is crucial for practitioners as it provides a more rigorous evaluation of AI's mathematical reasoning abilities, guiding future developments in model training and architecture.