In-context image editing on MICE-Bench (Turn 1)
66.31IFGPT-Image-1 [high]
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| GPT-Image-1 [high]Tier=Closed-source2026.06 | 66.31 | 75.22 | 61.29 | 67.61 | |
| Nano-Banana-2Tier=Closed-source2026.06 | 62.65 | 87.38 | 59.97 | 70 | |
| Qwen-Image-Edit [2511]Originally support in-context editing=false2026.06 | 58.62 | 81.2 | 52.84 | 64.22 | |
| Edit-R22026.06 | 54.82 | 88.62 | 50.73 | 64.72 | |
| BAGELvariant=w/ IC-CoT2026.06 | 51.88 | 87.8 | 46.3 | 62 | |
| BAGEL2026.06 | 51.38 | 89.12 | 44.72 | 61.74 | |
| BAGELvariant=w/ think2026.06 | 45.9 | 88.39 | 42.37 | 58.89 | |
| OmniGen2026.06 | 44.44 | 56.84 | 38.11 | 46.46 | |
| Flux.1-Kontext [dev]Originally support in-context editing=false2026.06 | 43.21 | 73.25 | 38.28 | 51.58 | |
| VINCIE2026.06 | 34.77 | 71.51 | 31.67 | 45.98 |