Image Editing on RISEBench
32.8RISEGemini-2.5-Flash-Image
Evaluation results
| Method | Links | |
|---|---|---|
| Gemini-2.5-Flash-ImageType=proprietary2026.01 | 32.8 | |
| GPT-Image-1Type=proprietary2026.01 | 28.9 | |
| OursType=diffusion, post-training=Dual-GRPO2026.01 | 23.9 | |
| Ours (w/o GRPO)Type=diffusion, post-training=SFT-only2026.01 | 20.2 | |
| Seedream-4.0Type=proprietary2026.01 | 10.8 | |
| Bagel w/ CoTType=unified2026.01 | 9.2 | |
| Qwen-Image-EditType=diffusion2026.01 | 8.9 | |
| FLUX.1-KontextType=diffusion2026.01 | 5.8 |