Integrated Multimodal Reasoning on OmniBench (held-out)
55.2Causal ScoreShiftDC
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| ShiftDCBase Model=Qwen3-Omni-30B-A3B-Instruct2026.05 | 55.2 | 40.5 | 34.5 | 49.2 | |
| BaseBase Model=Qwen3-Omni-30B-A3B-Instruct2026.05 | 54.9 | 40.5 | 31 | 48.8 | |
| ReGapBase Model=Qwen3-Omni-30B-A3B-Instruct2026.05 | 54.6 | 41.1 | 44.8 | 49.6 | |
| ReGapBase Model=MiniCPM-o-4.52026.05 | 50.7 | 50.3 | 27.6 | 49.2 | |
| BaseBase Model=MiniCPM-o-4.52026.05 | 50 | 47.2 | 27.6 | 47.8 | |
| ShiftDCBase Model=MiniCPM-o-4.52026.05 | 49.4 | 47.2 | 31 | 47.6 | |
| VLGuard-SFTBase Model=Qwen2.5-Omni-3B2026.05 | 43.2 | 52.2 | 34.5 | 45.6 | |
| ReGapBase Model=Qwen2.5-Omni-3B2026.05 | 31.8 | 42.9 | 20.7 | 34.8 | |
| BaseBase Model=Qwen2.5-Omni-3B2026.05 | 28.6 | 42.3 | 17.2 | 32.4 | |
| SPA-VL-DPOBase Model=Qwen2.5-Omni-3B2026.05 | 26.3 | 30.7 | 6.9 | 26.6 | |
| ShiftDCBase Model=Qwen2.5-Omni-3B2026.05 | 25.3 | 33.7 | 24.1 | 28 |