Paper 2605.13301
Achieving Gold-Medal-Level Olympiad Reasoning via Simple and Unified Scaling
- Published
- May 2026
- Research lab
- Independent
- Citations
- 4
- GitHub
- 94 stars
01 In brief
Summary
The paper introduces SU-01, a 30B-A3B reasoning model achieving gold-medal-level performance on mathematical and physical olympiads through a simple, unified post-training recipe.
The recipe comprises three stages: (1) SFT with a reverse-perplexity curriculum on 338K rigorous proof trajectories to instill proof-search and self-checking behaviors; (2) a two-level RL pipeline starting with coarse RL using verifiable rewards (GSPO) and progressing to refined RL with a generative proof reward (DeepSeekMath-V2), self-refinement prompts, and experience replay; and (3) test-time scaling via a self-verification-and-refinement loop.
SU-01 reaches 57.6% on IMO-ProofBench (70.2% with TTS), matches the strongest similar-size baseline on verifiable benchmarks (77.3% average), and scores 35 points on both IMO 2025 and USAMO 2026, meeting or exceeding gold lines.
It also shows transfer to scientific domains beyond math and physics.
The paper analyzes progressive reasoning gains, inference scaling (trajectories over 100K tokens), and the effectiveness of the reverse-perplexity ordering, highlighting the recipe's efficiency compared to larger-scale post-training efforts.
02 From the paper
Abstract
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.