| --- |
| dataset_info: |
| features: |
| - name: id |
| dtype: string |
| - name: task |
| dtype: string |
| - name: prompt |
| dtype: string |
| - name: image_rubrics |
| sequence: string |
| - name: text_rubrics |
| sequence: string |
| - name: image_ref |
| sequence: image |
| - name: text_ref |
| dtype: string |
| splits: |
| - name: test |
| num_bytes: 301995195.0 |
| num_examples: 1000 |
| download_size: 292190640 |
| dataset_size: 301995195.0 |
| configs: |
| - config_name: default |
| data_files: |
| - split: test |
| path: data/test-* |
| --- |
| # UEval: A Benchmark for Unified Multimodal Generation |
|
|
| > [**UEval: A Benchmark for Unified Multimodal Generation**](https://arxiv.org/abs/2502.12150) </br> |
| > *[Bo Li](https://primerl.github.io/), [Yida Yin](https://davidyyd.github.io), [Wenhao Chai](https://wenhaochai.com/), [Xingyu Fu](https://zeyofu.github.io/)\*, [Zhuang Liu](https://liuzhuang13.github.io)\* (* indicates co-advising) <br> |
| > Princeton University<br> |
| > [[Paper]](https://arxiv.org/abs/2601.22155) [[Project page]](https://zlab-princeton.github.io/UEval/) [[Code]](https://github.com/zlab-princeton/UEval) |
|
|
| --- |
| <p align="center"> |
| <img src="https://github.com/user-attachments/assets/f66379a8-c571-4cba-8e9b-78d158ecd26c" width=100% height=100% |
| class="center"> |
| </p> |
|
|
| We introduce **UEval**, a benchmark to evaluate unified models, i.e., models capable of generating both images and text. UEval comprises 1,000 expert-curated prompts that require both images and text in the model outputs, sourced from 8 diverse real-world domains. |
|
|
| ## Results |
|
|
| We evaluate recent unified models on all 8 tasks in our benchmark. Overall, frontier models consistently outperform open-source ones across all tasks: GPT-5-Thinking achieves the highest average score of 66.4, while the best open-source model obtains only 49.1. The gap between proprietary and open-source models is very large: the strongest frontier model (e.g., GPT-5-Thinking) outperforms the best open-source model (e.g., Emu 3.5) by over 17 points on average. |
|
|
| | Model | Space | Textbook | Diagram | Paper | Art | Life | Tech | Exercise | Avg | |
| | ------------------------------- | -------- | -------- | -------- | -------- | -------- | -------- | -------- | -------- | -------- | |
| | *Reference* | 96.2 | 94.4 | 93.1 | 96.2 | 90.6 | 87.7 | 90.6 | 89.2 | 92.2 | |
| | Janus-Pro | 21.0 | 31.0 | 37.4 | 15.2 | 26.4 | 23.0 | 17.6 | 11.5 | 22.9 | |
| | Show-o2 | 25.4 | 33.1 | 33.2 | 17.4 | 25.6 | 15.6 | 17.4 | 13.1 | 22.6 | |
| | MMaDA | 10.8 | 20.0 | 14.2 | 13.3 | 15.7 | 15.8 | 12.4 | 12.6 | 14.4 | |
| | BAGEL | 29.8 | 42.5 | 37.2 | 20.0 | 39.0 | 33.6 | 24.8 | 21.4 | 31.0 | |
| | Emu3.5 | **59.1** | **57.4** | **41.1** | **31.6** | **59.3** | **62.0** | **37.0** | **45.4** | **49.1** | |
| | Gemini-2.0-Flash | 65.2 | 55.2 | 47.6 | 45.8 | **70.4** | 58.0 | 50.2 | 48.0 | 55.1 | |
| | Gemini-2.5-Flash | 78.0 | 74.0 | 66.4 | **71.6** | 66.6 | 63.0 | **58.2** | 50.0 | 66.0 | |
| | GPT-5-Instant | 77.3 | 77.9 | 62.3 | 55.1 | 71.2 | **69.7** | 50.7 | 57.6 | 65.2 | |
| | GPT-5-Thinking | **84.0** | **78.0** | **67.8** | 51.9 | 67.8 | 63.8 | 57.0 | **61.4** | **66.4** | |
|
|
|
|
|
|
| <img width="1368" height="1417" alt="Image" src="https://github.com/user-attachments/assets/ede55cbf-fffd-4278-8fed-34d8c95596a8" /> |
|
|
| <img width="1398" height="1542" alt="Image" src="https://github.com/user-attachments/assets/add410df-b267-4693-8a3d-b9c8e45108fb" /> |
|
|
| ## Citation |
| If you find this repository helpful, please consider citing: |
| ```bibtex |
| @article{li2026ueval, |
| title = {UEval: A Benchmark for Unified Multimodal Generation}, |
| author = {Li, Bo and Yin, Yida and Chai, Wenhao and Fu, Xingyu and Liu, Zhuang}, |
| journal = {arXiv preprint arXiv:2601.22155}, |
| year = {2026} |
| } |
| ``` |
|
|
