𝗣𝗵𝘆𝘀𝗶𝗰𝘀 is one of the sharpest testbeds when reasoning models move from puzzle-solving to realistic scientific reasoning.
We have trained 𝗣𝟭 – a serie of models that achieved 𝗴𝗼𝗹𝗱 𝗺𝗲𝗱𝗮𝗹-𝗹𝗲𝘃𝗲𝗹 𝗽𝗲𝗿𝗳𝗼𝗿𝗺𝗮𝗻𝗰𝗲 in the Physics Olympiad.🥇
2
The flagship model P1-235B-A22B wins 12 golds and 1 silver across 13 top international/regional physics contests, tying Gemini-2.5-Pro for first on the medal table, and becomes the first open-source model to win gold at IPhO 2025.
3
The proposed co-evolution multi-agent system, PhysicsMinions, further boosts P1 to No.1 on both IPhO 2025 and HiPhO overall scores, outperforming Gemini-2.5-Pro and GPT-5.
4
HiPhO compiles 13 olympiad-level contests from 2024–2025 including IPhO, APhO, and EuPhO. Using official rubrics, it scores answers and processes with fine-grained, human-aligned criteria.
With this multi-stage RL strategy, P1 achieves sustained, long-term improvements over the base language model.
6
Beyond physics reasoning, P1 improves across multiple domains, including math, coding, and STEM benchmarks, demonstrating strong generalization of physics reasoning.
7
Case study.
8
We thank @Alibaba_Qwen @Zai_org (slime) @lmsysorg @verl_project @sgl_project for their strong base and infrastructure!