Instruction Synthesis for Image Editing on ByteMorph-Bench
4.588S ScoreQwen3-VL-235B-A22B-Instruct
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Qwen3-VL-235B-A22B-InstructTraining=SFT+DPO2026.04 | 4.588 | 4.7 | 4.75 | 4.04 | |
| Gemini-3-Pro2026.04 | 4.522 | 4.55 | 4.7 | 4.11 | |
| Qwen3-VL-235B-A22B-InstructTraining=SFT2026.04 | 4.208 | 4.07 | 4.51 | 3.89 | |
| Qwen3-VL-32B-InstructTraining=SFT+DPO2026.04 | 3.931 | 3.73 | 4.17 | 3.86 | |
| Qwen3-VL-32B-InstructTraining=SFT2026.04 | 3.914 | 3.78 | 4.08 | 3.84 | |
| Qwen3.5-397B-A17B2026.04 | 3.867 | 3.71 | 4.16 | 3.61 | |
| Kimi-K2.52026.04 | 3.679 | 2.94 | 4.57 | 3.38 | |
| Qwen3-VL-235B-A22B-Instruct2026.04 | 3.462 | 3.19 | 3.9 | 3.13 | |
| GLM-4.5V (106B-A12B)2026.04 | 3.448 | 3.25 | 3.78 | 3.19 | |
| GPT-4.12026.04 | 3.412 | 3.27 | 3.68 | 3.14 | |
| Qwen3-VL-32B-Instruct2026.04 | 3.332 | 3.06 | 3.67 | 3.2 |