Instruction Synthesis for Image Editing on Eval-400 In-house
4.712S ScoreQwen3-VL-235B-A22B-Instruct
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Qwen3-VL-235B-A22B-InstructTraining=SFT+DPO2026.04 | 4.712 | 4.71 | 4.87 | 4.4 | |
| Gemini-3-Pro2026.04 | 4.706 | 4.7 | 4.85 | 4.43 | |
| Qwen3-VL-235B-A22B-InstructTraining=SFT2026.04 | 4.521 | 4.44 | 4.79 | 4.15 | |
| Qwen3-VL-32B-InstructTraining=SFT+DPO2026.04 | 4.386 | 4.31 | 4.63 | 4.05 | |
| Qwen3.5-397B-A17B2026.04 | 4.38 | 4.3 | 4.62 | 4.06 | |
| Qwen3-VL-32B-InstructTraining=SFT2026.04 | 4.349 | 4.3 | 4.58 | 3.98 | |
| GPT-4.12026.04 | 4.22 | 4.03 | 4.6 | 3.84 | |
| Kimi-K2.52026.04 | 4.111 | 3.69 | 4.72 | 3.74 | |
| GLM-4.5V (106B-A12B)2026.04 | 3.97 | 4.36 | 4.33 | 3.52 | |
| Qwen3-VL-235B-A22B-Instruct2026.04 | 3.88 | 3.65 | 3.78 | 3.48 | |
| Qwen3-VL-32B-Instruct2026.04 | 3.48 | 3.14 | 3.6 | 2.92 |