---
license: apache-2.0
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
- config_name: dev
  data_files:
  - split: total
    path: dev/total-*
  - split: dev
    path: dev/dev-*
dataset_info:
- config_name: default
  features:
  - name: index
    dtype: int64
  - name: question
    dtype: string
  - name: answer
    dtype: string
  - name: images
    list: image
  - name: reasoning
    dtype: string
  - name: sig_figs
    dtype: string
  - name: level
    dtype: int64
  - name: subject
    dtype: string
  - name: language
    dtype: string
  - name: img_category
    dtype: string
  - name: vision_relevance
    dtype: string
  - name: caption
    dtype: string
  splits:
  - name: train
    num_bytes: 90258908.0
    num_examples: 2000
  download_size: 77879212
  dataset_size: 90258908.0
- config_name: dev
  features:
  - name: question
    dtype: string
  - name: subject
    dtype: string
  - name: image_path
    sequence: string
  - name: sig_figs
    dtype: string
  - name: level
    dtype: int64
  - name: language
    dtype: string
  - name: index
    dtype: int64
  - name: img_category
    dtype: string
  - name: vision_relevance
    dtype: string
  - name: caption
    dtype: string
  - name: image_0
    dtype: image
  - name: image_1
    dtype: image
  - name: image_2
    dtype: image
  - name: image_3
    dtype: image
  splits:
  - name: total
    num_bytes: 96133884.0
    num_examples: 2000
  - name: dev
    num_bytes: 9343791.0
    num_examples: 200
  download_size: 86916417
  dataset_size: 105477675.0
task_categories:
- question-answering
- visual-question-answering
language:
- en
tags:
- physics
- multi-modal
size_categories:
- 1K<n<10K
---
# SeePhys: Does Seeing Help Thinking? -- Benchmarking Vision-Based Physics Reasoning
Can AI truly see the Physics? Test your model with the newly released SeePhys Benchmark!
Covering 2,000 vision-text multimodal physics problems spanning from middle school to doctoral qualification exams, the SeePhys benchmark systematically evaluates LLMs/MLLMs on tasks integrating complex scientific diagrams with theoretical derivations. Experiments reveal that even SOTA models like Gemini-2.5-Pro and o4-mini achieve accuracy rates below 55%, with over 30% error rates on simple middle-school-level problems, highlighting significant challenges in multimodal reasoning.

The benchmark is now open for evaluation at the ICML 2025 AI for MATH Workshop. Academic and industrial teams are invited to test their models!

🔗 Key Links:

📜Paper: http://arxiv.org/abs/2505.19099

⚛️Project Page: https://seephys.github.io/

🏆Challenge Submission: https://www.codabench.org/competitions/7925/

➡️Competition Guidelines: https://sites.google.com/view/ai4mathworkshopicml2025/challenge


The answer will be announced on July 1st, 2025 (Anywhere on Earth, AoE), which is after the submission deadline for the ICML 2025 Challenges on Automated Math Reasoning and Extensions.

If you find SeePhys useful for your research and applications, please kindly cite using this BibTeX:

```
@article{xiang2025seephys,
  title={SeePhys: Does Seeing Help Thinking? -- Benchmarking Vision-Based Physics Reasoning},
  author={Kun Xiang, Heng Li, Terry Jingchen Zhang, Yinya Huang, Zirong Liu, Peixin Qu, Jixi He, Jiaqi Chen, Yu-Jie Yuan, Jianhua Han, Hang Xu, Hanhui Li, Mrinmaya Sachan, Xiaodan Liang},
  journal={arXiv preprint arXiv:2505.19099},
  year={2025}
}
```