Achieving Gold-Medal-Level Olympiad Reasoning via Simple and Unified Scaling
Yafu Li, Runzhe Zhan, Haoran Zhang, Shunkai Zhang, Yizhuo Li, Zhilin Wang, Jiacheng Chen, Futing Wang, Xuyang Hu, Yuchen Fan, Bangjie Xu, Yucheng Su, Xinmiao Han, Chenxi Li, Haodi Lei, Yufeng Zhao, Zejin Lin, Qianjia Cheng, Tong Zhu, Xiaoye Qu, Ganqu Cui, Peng Ye, Yun Luo, Zhouchen Lin, Yu Qiao, Bowen Zhou, Ning Ding, Yu Cheng
arXiv:2605.13301 · 165▲ · cs.AI, cs.CL
View on arXiv →Premium readers get an interactive explainer for this paper — a figure you can poke at, not just read.
What it is
The paper describes a training recipe that turns a 30B-A3B post-trained reasoning model into an olympiad-level math and physics solver, using three stages: supervised fine-tuning ordered by reverse-perplexity curriculum, a two-stage RL pipeline (verifiable-reward RL followed by proof-quality RL with a generative reward model, self-refinement, and experience replay), and a test-time solve-verify-refine loop. The resulting model, SU-01, is trained on 338K sub-8K-token SFT trajectories plus 200 RL steps.
Why it matters
It shows a compact 30B mixture-of-experts model can reach performance close to much larger frontier systems on proof-heavy tasks, mostly by spending inference compute on iterative self-verification rather than raw model size. For practitioners, the interesting lever is the test-time refinement loop, which improved proof-grading scores substantially without changing the model.
Practical takeaway
If you are building reasoning systems, expect the biggest gains on rigorous multi-step tasks to come from a structured propose-verify-refine loop at inference rather than single-pass sampling, and from ordering SFT data by perplexity (hardest-to-model examples first) to reshape behavior without wrecking existing capability.
Key result
On IMO-ProofBench, SU-01 scored 57.6% with direct generation and 70.2% with test-time scaling, versus 23.1% for the similar-size Qwen3.6-35B-A3B baseline. Caveat: the headline competition results (35/35 gold-level points on IMO 2025 and USAMO 2026) were graded by human experts only in the TTS condition and rest on a handful of problems, and those TTS runs sustain 100K+ token trajectories, so the compute cost per problem is high.
Jargon
Subscribe
Get the next issue.
Free. One email a week. Unsubscribe any time: no account, no dark patterns.