Achieving Gold-Medal-Level Olympiad Reasoning via Simple and Unified Scaling
This work provides a practical, scalable recipe for achieving top-tier olympiad-level reasoning in AI, potentially advancing automated scientific problem solving.
The authors present a simple and unified recipe combining reverse-perplexity curriculum SFT and two-stage RL to train a 30B-A3B model (SU-01) that achieves gold-medal-level performance on IMO 2025/USAMO 2026 and IPhO 2024/2025, with strong generalization to other scientific domains.
Recent progress in reasoning models has substantially advanced long-horizon mathematical and scientific problem solving, with several systems now reaching gold-medal-level performance on International Mathematical Olympiad (IMO) and International Physics Olympiad (IPhO) problems. In this paper, we introduce a simple and unified recipe for converting a post-trained reasoning backbone into a rigorous olympiad-level solver. The recipe first uses a reverse-perplexity curriculum for SFT to instill rigorous proof-search and self-checking behaviors, then scales these behaviors through a two-stage RL pipeline that progresses from RL with verifiable rewards to more delicate proof-level RL, and finally boosts solving performance with test-time scaling. Applying this recipe, we train a 30B-A3B backbone with SFT on around 340K sub-8K-token trajectories followed by 200 RL steps. The resulting model, SU-01, supports stable reasoning on difficult problems with trajectories exceeding 100K tokens, while achieving gold-medal-level performance on mathematical and physical olympiad competitions, including IMO 2025/USAMO 2026 and IPhO 2024/2025. It also demonstrates strong generalization of scientific reasoning to domains beyond mathematics and physics.