Multimodal Reasoning on R1-Onevision-Bench Deduction
27.8AccuracyNo Compression
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| No CompressionBase Model=Qwen2.5-VL2026.02 | 27.8 | 834.3 | 30 | |
| XMCCBase Model=Qwen2.5-VL2026.02 | 27.6 | 115 | 4.2 | |
| Prune-on-LogicBase Model=Qwen2.5-VL2026.02 | 27.3 | 588.3 | 21.5 | |
| StepEntropyBase Model=Qwen2.5-VL2026.02 | 26.5 | 522.2 | 19.7 | |
| Prune-on-LogicBase Model=Qwen3-VL2026.02 | 23 | 602.6 | 26.2 | |
| XMCCBase Model=Qwen3-VL2026.02 | 22.8 | 103.7 | 4.5 | |
| StepEntropyBase Model=Qwen3-VL2026.02 | 22.5 | 560 | 24.9 | |
| No CompressionBase Model=Qwen3-VL2026.02 | 21.4 | 819 | 38.3 |