Knowledge-intensive image editing on UniREditBench (test)
81.01Real World ScoreGPT-4o
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| GPT-4omodel_source=closed-source2026.02 | 81.01 | 62.07 | 73.39 | |
| UniReasonreasoning_protocol=w/ Reasoning, reasoning_type=interleaved reasoning and refinement2026.02 | 74.82 | 65.3 | 70.06 | |
| Qwen-Image-Editreasoning_protocol=w/o Reasoning2026.02 | 70.95 | 41.92 | 56.52 | |
| Seedream 4.0model_source=closed-source2026.02 | 66.22 | 45.38 | 55.77 | |
| BAGELreasoning_protocol=w/ Reasoning, reasoning_type=textual reasoning only2026.02 | 56.8 | 45.1 | 50.96 | |
| OmniGen2reasoning_protocol=w/o Reasoning2026.02 | 53.69 | 33.14 | 43.41 | |
| Lumina-DiMOOreasoning_protocol=w/o Reasoning2026.02 | 51.44 | 45.61 | 48.54 |