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.
Mastering Olympiad-Level Physics with Artificial Intelligence
School of Physics, Peking University(物理系,北京大学);School of Electronics Engineering and Computer Science, Peking University(电子工程与计算机科学系,北京大学);Center for High Energy Physics, Peking University(高能物理中心,北京大学)
我们提出了DrQ-v2,一个无模型强化学习(RL)算法的视觉连续控制。DrQ-v2建立在DrQ的基础上,DrQ是一种非策略的行动者-批评者方法,使用数据增强直接从像素中学习。我们介绍了几项改进,这些改进在DeepMind Control Suite上产生了最先进的结果。值得注意的是,DrQ-v2能够直接从像素观察中解决复杂的人形运动任务,这是以前无模型RL无法实现的。DrQ-v2概念简单,易于实现,与以前的工作相比,它提供了更好的计算占用空间,大多数任务只需8小时即可在单个GPU上进行训练。最后,我们公开发布DrQ-v2的实现,为RL从业者提供强大且计算效率高的基线。
School of Artificial Intelligence, Jilin University(吉林大学人工智能学院);Sun Yat-sen University(中山大学);Lehigh University(里海大学);College of Computing and Data Science, NTU(南洋理工大学计算与数据科学学院)
Suphx: Mastering Mahjong with Deep Reinforcement Learning
Microsoft Research Asia(微软亚洲研究院);Kyoto University(京都大学);University of Science and Technology of China(中国科学技术大学);Tsinghua University(清华大学);Nankai University(南开大学)
Beyond Reasoning: RL-Policy Guided LLM Inference for Efficient Strategy in Liar's Poker
Allometry Labs(Allometry实验室);Graduate School of Business, Columbia University(哥伦比亚大学商学院);Sauder School of Business, University of British Columbia(不列颠哥伦比亚大学萨德尔商学院)
Mastering Strategy Card Game (Legends of Code and Magic) via End-to-End Policy and Optimistic Smooth Fictitious Play
ByteDance(字节跳动)
深度强化学习结合Fictionary Play在许多基准测试游戏上显示出令人印象深刻的结果,然而,其中大多数都是单阶段的。相比之下,现实世界的决策问题可能由多个阶段组成,其中观察空间和行动空间在各个阶段之间可能完全不同。我们研究了一个两阶段的策略纸牌游戏Legends of Code and Magic,并提出了一个端到端的策略来解决多阶段游戏中出现的困难。我们还提出了一个乐观光滑虚拟博弈算法来寻找两人博弈的纳什均衡。我们的方法赢得了COG2022比赛的双重冠军。大量的研究验证并显示了我们方法的先进性。
LLaMA-Berry: Pairwise Optimization for O1-like Olympiad-Level Mathematical Reasoning
Fudan University(复旦大学);Shanghai Artificial Intelligence Laboratory(上海人工智能实验室);Hong Kong Polytechnic University(香港理工大学);University of New South Wales(新南威尔士大学);Shanghai Jiao Tong University(上海交通大学);Stanford University(斯坦福大学);University of California, Merced(加州大学默塞德分校)
Mastering Continual Reinforcement Learning through Fine-Grained Sparse Network Allocation and Dormant Neuron Exploration
IHPC, Agency for Science, Technology and Research (A*STAR)(新加坡科技研究局IHPC);College of Computing and Data Science, Nanyang Technological University (NTU)(南洋理工大学计算与数据科学学院)
Doctor-R1: Mastering Clinical Inquiry with Experiential Agentic Reinforcement Learning
Dept. of Comp. Sci. & Tech., Institute for AI, Tsinghua University(计算机科学与技术系,人工智能研究院,清华大学);College of AI, Tsinghua University(人工智能学院,清华大学);Institute for AI Industry Research (AIR), Tsinghua University(人工智能产业研究院(AIR),清华大学)
多智能体足球是人工智能研究中一个尚未解决的挑战。现有的工作集中在解决游戏的简化场景,或者利用专家演示。在本文中,我们开发了一个多智能体系统来玩完整的11对11游戏模式,没有演示。这种游戏模式包含了对现代强化学习算法的主要挑战;多智能体协调,长期规划和非传递性。为了应对这些挑战,我们提出了TiZero;一个自我进化的多智能体系统,从头开始学习。TiZero引入了几项创新,包括自适应课程学习,一种新颖的自我游戏策略,以及一个共同优化多个代理策略的目标。在实验中,它在Google Research Football环境中的表现远远优于以前的系统,将胜率提高了30%以上。为了展示TiZero的创新的普遍性,他们在足球之外的几个环境中进行了评估;过度烹饪,多智能体粒子环境,井字游戏和连接四。
EfficientZero V2: Mastering Discrete and Continuous Control with Limited Data
Institute for Interdisciplinary Information Sciences, Tsinghua University(清华大学交叉信息研究院);Shanghai Qi Zhi Institute(上海期智研究院);Shanghai Artificial Intelligence Laboratory(上海人工智能实验室);Texas A&M University(德克萨斯农工大学)
DrM: Mastering Visual Reinforcement Learning through Dormant Ratio Minimization
Tsinghua University(清华大学);University of Maryland, College Park(马里兰大学帕克分校);Shanghai Jiao Tong University(上海交通大学);Shanghai Qi Zhi Institute(上海期智研究院);Shanghai AI Lab(上海人工智能实验室)
Intern-S1-MO: Long-horizon Reasoning Agent for Olympiad?Level Mathematical Problem Solving
Shanghai AI Laboratory(上海人工智能实验室);Shanghai Jiao Tong University(上海交通大学);MMLab;The Chinese University of Hong Kong(香港中文大学);ICMAT Spanish National Research Council(西班牙国家科研 council);The High School Affiliated to Renmin University of China(中国人民大学附属高中);Ren Hui Academy of Beijing(北京润辉学院)