Image Captioning on CAPability
75.7Average ScoreGemini-3-Pro-Preview
Evaluation Results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Gemini-3-Pro-Preview2026.05 | 75.7 | 83.3 | 48.5 | 72.8 | 62.8 | 80.3 | 76.6 | 96.8 | 91 | 69.2 | |
| CycleCapT2I Model=FLUX.1-dev2026.03 | 73.73 | — | — | — | — | — | — | — | — | — | |
| CycleCapT2I Model=SD32026.03 | 72.95 | — | — | — | — | — | — | — | — | — | |
| CLAIMDIFF-RLReward Configuration=relative, Initialization=SFT2026.05 | 71.5 | 82.8 | 49.8 | 71.7 | 64.2 | 81.1 | 67 | 95.4 | 88.1 | 43.1 | |
| CyclePrefT2I Model=SD32026.03 | 70.59 | — | — | — | — | — | — | — | — | — | |
| Baseline (Qwen2-VL-7B)T2I Model=—2026.03 | 70.47 | — | — | — | — | — | — | — | — | — | |
| SFT2026.05 | 69.5 | 81.2 | 44.1 | 71 | 57.9 | 79 | 63.8 | 95.4 | 88.6 | 44.5 | |
| CLAIMDIFF-RLReward Configuration=actor-only, Initialization=SFT2026.05 | 69.3 | 79.5 | 45.6 | 72.1 | 59 | 79.3 | 65.5 | 93 | 86 | 44 | |
| Qwen3VL-32B-Instruct2026.05 | 69 | 80.7 | 43.5 | 74.4 | 58.9 | 80.2 | 55.1 | 95 | 91.9 | 41.9 | |
| RICO-FlashT2I Model=FLUX.1-dev2026.03 | 68.83 | — | — | — | — | — | — | — | — | — | |
| HOLISTIC-RLReward Configuration=w/ ref, Initialization=SFT2026.05 | 67.9 | 79.8 | 37 | 70.7 | 57.2 | 78.3 | 66.3 | 93.8 | 89.1 | 38.9 | |
| HOLISTIC-RLReward Configuration=w/o ref, Initialization=SFT2026.05 | 65.8 | 74.8 | 31.1 | 67.8 | 55.6 | 78.2 | 67.3 | 91.5 | 88.7 | 37.4 |