Paper 2602.09443
P1-VL: Bridging Visual Perception and Scientific Reasoning in Physics Olympiads
- Published
- Feb 2026
- Research lab
- Independent
- Citations
- 5
- GitHub
- 15 stars
01 In brief
Summary
The P1-VL technical report introduces a family of open-source vision-language models (VLMs) designed for advanced scientific reasoning, specifically targeting physics Olympiad problems.
The models are trained exclusively via reinforcement learning (RL) using a curriculum that progressively increases problem difficulty and expands exploration space, combined with a masked importance sampling technique to stabilize training.
At inference, the models are augmented with the PhysicsMinions agent framework for iterative self-verification.
The flagship model, P1-VL-235B-A22B, achieves 12 gold medals on the HiPhO benchmark (13 exams from 2024–2025), ranking 3rd overall among all models and 1st among open-source models.
With agentic augmentation, it ranks 2nd globally, surpassing GPT-5.2.
A smaller variant, P1-VL-30B-A3B, also performs well, achieving 9 gold medals.
The models demonstrate strong generalization to other STEM benchmarks, outperforming their base models in mathematics and multimodal reasoning tasks.
The training data consists of 8,033 curated physics problems from Olympiads and textbooks, with rigorous quality control.
The report highlights the importance of visual perception in physics reasoning, as diagrams often contain essential constraints.
02 From the paper
Abstract
The transition from symbolic manipulation to science-grade reasoning represents a pivotal frontier for Large Language Models (LLMs), with physics serving as the critical test anchor for binding abstract logic to physical reality. Physics demands that a model maintain physical consistency with the laws governing the universe, a task that fundamentally requires multimodal perception to ground abstract logic in reality. At the Olympiad level, diagrams are often constitutive rather than illustrative, containing essential constraints, such as boundary conditions and spatial symmetries, that are absent from the text. To bridge this visual-logical gap, we introduce P1-VL, a family of open-source vision-language models engineered for advanced scientific reasoning. Our method harmonizes Curriculum Reinforcement Learning, which employs progressive difficulty expansion to stabilize post-training, with Agentic Augmentation, enabling iterative self-verification at inference. Evaluated on HiPhO, a rigorous benchmark of 13 exams from 2024-2025, our flagship P1-VL-235B-A22B becomes the first open-source Vision-Language Model (VLM) to secure 12 gold medals and achieves the state-of-the-art performance in the open-source models. Our agent-augmented system achieves the No.2 overall rank globally, trailing only Gemini-3-Pro. Beyond physics, P1-VL demonstrates remarkable scientific reasoning capacity and generalizability, establishing significant leads over base models in STEM benchmarks. By open-sourcing P1-VL, we provide a foundational step toward general-purpose physical intelligence to better align visual perceptions with abstract physical laws for machine scientific discovery.