OmniFysics-Nano-V2: Understanding the Physical World Across Modalities

📄 Paper • 🌐 Project Page • 🤗 Model • 📚 Citation

We present OmniFysics‑Nano‑V2, a compact 4B omni‑modal model that supports holistic understanding of images, videos, audios, speeches, and texts with speech output capability. We design a dual‑branch physical supervision pipeline that complements static physical attribute grounding and dynamic physical event modeling. We develop a reward‑diversity filtering for policy‑aware RL data curation, and a two‑stage GRPO strategy that optimizes general task performance and enhances fine‑grained physical reasoning via intermediate perception supervision. The proposed model achieves leading result on 17 of 21 Benchmarks against SOTA omni‑modal models.

Architecture of OmniFysics-Nano-V2

We propose the OmniFysics-Nano-V2, a compact 4B omni-modal model for physical-world perception and understanding. The model supports the understanding of image, video, audio, speech, and text inputs, alongside text and audio generation capabilities.

To address the problem of Absent Physical Supervision, we construct a physics-aware data pipeline with complementary Static and Dynamic branches.

We develop a reward-diversity filtering for policy-aware RL data curation, and a two-stage GRPO strategy that optimizes general task performance and enhances fine-grained physical reasoning via intermediate perception supervision.

Performance

We evaluated OmniFysics-Nano-V2 on 21 Benchmarks that cover general multimodal, audio, omni-modal / video, physical understanding, mathematical reasoning and physical reasoning benchmarks. The OmniFysics-Nano-V2 achieves SOTA performance on 17 Benchmarks with 4B model size, even against 7B-scale baseline models. Notably, the OmniFysics-Nano-V2 achieves 98.27% on FysicsEval-Understanding and 59.42% on PhysUniBench, surpassing the SOTA baselines by 5.57% and 11.42%, respectively.

• Comprehensive Physics Benchmarks

  

• Omni-Modal and Video Benchmarks

  

• Visual Multimodal Understanding Benchmarks

  

• Audio Understanding Benchmarks

Acknowledgments

This work is built upon the following open‑source repositories. We sincerely thank the authors and contributors for releasing their code.

• VeOmni: for the foundational model backbone implementation.
• CosyVoice3: for speech generation.
• ms-swift: for reinforcement learning training.

Citation

@article{liu2026omnifysics-nano-v2,
  title   = {OmniFysics-Nano-V2 Technical Report: Understanding the Physical World Across Modalities},
  author  = {Yizhou Liu, Jinghang Han, Kaixiang Qiu, Qi He, Minghao Han, Yue Jiang, Xujia Chen, Wei Zou, Shunli Wang, Lihua Zhang, Dingkang Yang},
  journal = {arXiv preprint},
  eprint={2609.25738},
  url={https://arxiv.org/abs/2609.25738}, 
  primaryClass={cs.AI},
  year    = {2026}
}

License

The content of this repository is released under the Apache License 2.0 with an additional non-commercial restriction: it may be used, reproduced, and distributed for research and educational purposes only. Any commercial use is prohibited without prior written permission from the maintainers. Source videos remain subject to the licenses of their original datasets.