Tag
This study evaluates frontier and open-weight LLMs on IMO 2026 problems, demonstrating that specialized harnesses like AutoFyn significantly improve performance of sub-frontier models, though hallucination issues persist on the hardest problem.