Reasoning Logic Image Generation on RISEBench
77Instr. Reas.Nano Banana Pro
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Nano Banana ProAccess=Proprietary2026.02 | 77 | 85.5 | 94.4 | 47.2 | |
| GPT-Image-1Access=Proprietary2026.02 | 62.8 | 80.2 | 94.9 | 28.9 | |
| Mind-BrushAccess=Open-source2026.02 | 61.5 | 79.4 | 86.5 | 24.7 | |
| Nano BananaAccess=Proprietary2026.02 | 61.2 | 86 | 91.3 | 32.8 | |
| Qwen-ImageAccess=Open-source2026.02 | 49.9 | 71 | 91.5 | 19.4 | |
| BAGELAccess=Open-source, CoT=true2026.02 | 45.9 | 73.8 | 80.1 | 11.9 | |
| BAGELAccess=Open-source, CoT=false2026.02 | 36.5 | 53.5 | 73 | 6.1 | |
| FLUX.1-devAccess=Open-source2026.02 | 26 | 71.6 | 85.2 | 1.9 | |
| FLUX.1-CannyAccess=Open-source2026.02 | 20.2 | 13.1 | 77.5 | 0 |