Multimodal Reasoning on R1-Onevision-Bench Math
25.7AccuracyNo Compression
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| No CompressionBase Model=Qwen2.5-VL2026.02 | 25.7 | 629.5 | 24.5 | |
| Prune-on-LogicBase Model=Qwen2.5-VL2026.02 | 25.4 | 418.7 | 16.5 | |
| XMCCBase Model=Qwen2.5-VL2026.02 | 25.4 | 121.2 | 4.8 | |
| Prune-on-LogicBase Model=Qwen3-VL2026.02 | 25 | 428.9 | 17.2 | |
| StepEntropyBase Model=Qwen2.5-VL2026.02 | 24.9 | 399 | 16 | |
| No CompressionBase Model=Qwen3-VL2026.02 | 24.8 | 617.7 | 24.9 | |
| XMCCBase Model=Qwen3-VL2026.02 | 24.5 | 120.6 | 4.9 | |
| StepEntropyBase Model=Qwen3-VL2026.02 | 22.9 | 397.7 | 17.4 |