Paper 2511.13612
P1: Mastering Physics Olympiads with Reinforcement Learning
- Published
- Nov 2025
- Research lab
- Independent
- Citations
- 4
- GitHub
- 89 stars
01 In brief
Summary
The P1 technical report introduces a family of open-source physics reasoning models trained entirely via reinforcement learning (RL).
P1-235B-A22B is the first open-source model to achieve gold-medal performance at the International Physics Olympiad 2025 (IPhO 2025), scoring 21.2/30 and ranking 3rd behind Gemini-2.5-Pro and GPT-5.
It won 12 gold and 1 silver across 13 international/regional physics competitions in 2024/2025.
The smaller P1-30B-A3B also achieved silver at IPhO 2025, outperforming most open-source models.
Combined with the PhysicsMinions agent framework, P1-235B-A22B achieved the No.1 overall score on IPhO 2025 and the highest average across all 13 competitions.
The models were trained using a multi-stage RL framework with adaptive learnability adjustment and training stabilization, based on a curated dataset of 5,065 Olympiad-level physics problems.
P1 models also showed strong generalizability, outperforming their base models on math, coding, and general reasoning benchmarks.
The report also discusses the risks of using model-based verifiers during training, favoring rule-based verification for stability.
The P1 ecosystem is fully open-source, including models, training algorithms, evaluation benchmarks, and an agentic inference framework.
On CPhO 2025, P1-235B-A22B scored 227/320, exceeding the top-1 human medalist's 199.
The work represents a significant step toward LLMs capable of genuine scientific reasoning and future physics research assistance.
- P1-235B-A22B: first open-source model with gold-medal performance at IPhO 2025, scoring 21.2/30.
- P1-30B-A3B: achieved silver at IPhO 2025, outperforming most open-source models.
- With PhysicsMinions, P1-235B-A22B achieved No.1 on IPhO 2025 and highest average across 13 competitions.
- Training used a multi-stage RL framework with adaptive…
02 From the paper
Abstract
Recent progress in large language models (LLMs) has moved the frontier from puzzle-solving to science-grade reasoning-the kind needed to tackle problems whose answers must stand against nature, not merely fit a rubric. Physics is the sharpest test of this shift, which binds symbols to reality in a fundamental way, serving as the cornerstone of most modern technologies. In this work, we manage to advance physics research by developing large language models with exceptional physics reasoning capabilities, especially excel at solving Olympiad-level physics problems. We introduce P1, a family of open-source physics reasoning models trained entirely through reinforcement learning (RL). Among them, P1-235B-A22B is the first open-source model with Gold-medal performance at the latest International Physics Olympiad (IPhO 2025), and wins 12 gold medals out of 13 international/regional physics competitions in 2024/2025. P1-30B-A3B also surpasses almost all other open-source models on IPhO 2025, getting a silver medal. Further equipped with an agentic framework PhysicsMinions, P1-235B-A22B+PhysicsMinions achieves overall No.1 on IPhO 2025, and obtains the highest average score over the 13 physics competitions. Besides physics, P1 models also present great performance on other reasoning tasks like math and coding, showing the great generalibility of P1 series.