Multi-step Editing on BlenderBench
33.14PLQwen3-VL-8B
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Qwen3-VL-8BSetting=One-Shot2026.01 | 33.14 | 37.51 | 1.61 | |
| BlenderAlchemyBase Model=Qwen3-VL-8B, Setting=best-of-12026.01 | 10.69 | 13.92 | 2.64 | |
| GPT-4oSetting=One-Shot2026.01 | 7.36 | 7.12 | 2.75 | |
| BlenderAlchemyBase Model=GPT-4o, Setting=best-of-12026.01 | 6.13 | 7.71 | 3.08 | |
| BlenderAlchemyBase Model=Qwen3-VL-8B, Setting=best-of-42026.01 | 5.38 | 4.64 | 2.94 | |
| VIGABase Model=GPT-4o, Setting=best-of-12026.01 | 5.11 | 4.33 | 3.58 | |
| VIGABase Model=Qwen3-VL-8B, Setting=best-of-42026.01 | 5.02 | 3.86 | 3.08 | |
| VIGABase Model=Qwen3-VL-8B, Setting=best-of-12026.01 | 3.85 | 3.34 | 3.33 | |
| VIGABase Model=GPT-4o, Setting=best-of-42026.01 | 2.94 | 3.5 | 3.83 | |
| BlenderAlchemyBase Model=GPT-4o, Setting=best-of-42026.01 | 1.95 | 2.47 | 3.53 |