Instruction-following generation on GenEval++ (test)
90Color AccuracyGPT-4o
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| GPT-4oorganization=OpenAI2026.02 | 90 | 67.5 | 72.5 | 62.5 | 60 | 80 | 85 | 73.9 | |
| Echo-4o2026.02 | 80 | 57.5 | 55 | 77.5 | 62.5 | 80 | 62.5 | 67.9 | |
| T2I-R12026.02 | 67.5 | 32.5 | 20 | 35 | 7.5 | 25 | 30 | 31.1 | |
| BAGEL + Reason-Reflect-Refinestage=full (reasoning + 4 reflection-refinement stages)2026.02 | 67.5 | 72.5 | 57.5 | 72.5 | 75 | 57.5 | 80 | 68.9 | |
| BAGEL + Reason-Reflect-Refinestage=reasoning only2026.02 | 50 | 65 | 60 | 65 | 55 | 60 | 60 | 59.3 | |
| Janus-Pro2026.02 | 45 | 30 | 12.5 | 30 | 7.5 | 35 | 12.5 | 24.6 | |
| FLUX.1-Kontextorganization=Labs2026.02 | 42.5 | 50 | 20 | 25 | 30 | 40 | 32.5 | 34.3 | |
| FLUX.1-devorganization=Labs2026.02 | 35 | 62.5 | 15 | 27.5 | 20 | 37.5 | 22.5 | 31.4 | |
| BAGEL2026.02 | 32.5 | 60 | 25 | 32.5 | 25 | 47.5 | 37.5 | 37.1 |