Visual Reasoning on LogicVista
52.13AccuracyVanilla
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| VanillaBackbone=InternVL3.5-8B, Token Retention Ratio=Full Tokens, DSTP=false2026.04 | 52.13 | — | 100 | — | |
| VanillaBackbone=Qwen3-VL-4B, Token Retention Ratio=Full Tokens, DSTP=false2026.04 | 49.22 | — | 100 | — | |
| DSTPBackbone=InternVL3.5-8B, Token Retention Ratio=33.3%, Pruning Method=VisionZip (CVPR’25), DSTP=true2026.04 | 48.76 | 93.5 | 87.83 | — | |
| DSTPBackbone=InternVL3.5-8B, Token Retention Ratio=33.3%, Pruning Method=FastV (ECCV’24), DSTP=true2026.04 | 48.44 | 92.9 | 92.44 | — | |
| DSTPBackbone=InternVL3.5-8B, Token Retention Ratio=33.3%, Pruning Method=DivPrune (CVPR’25), DSTP=true2026.04 | 47.87 | 91.8 | 90.09 | — | |
| DSTPBackbone=InternVL3.5-8B, Token Retention Ratio=22.2%, Pruning Method=FastV (ECCV’24), DSTP=true2026.04 | 46.08 | 88.4 | 87 | — | |
| DSTPBackbone=InternVL3.5-8B, Token Retention Ratio=22.2%, Pruning Method=DivPrune (CVPR’25), DSTP=true2026.04 | 44.74 | 85.8 | 82.98 | — | |
| DSTPBackbone=InternVL3.5-8B, Token Retention Ratio=22.2%, Pruning Method=VisionZip (CVPR’25), DSTP=true2026.04 | 43.62 | 83.7 | 81.01 | — | |
| DSTPBackbone=Qwen3-VL-4B, Token Retention Ratio=33.3%, Pruning Method=DivPrune (CVPR’25), DSTP=true2026.04 | 42.7 | 86.8 | 83.8 | — | |
| DSTPBackbone=Qwen3-VL-4B, Token Retention Ratio=33.3%, Pruning Method=FastV (ECCV’24), DSTP=true2026.04 | 41.16 | 83.6 | 82.9 | — | |
| DSTPBackbone=Qwen3-VL-4B, Token Retention Ratio=33.3%, Pruning Method=VisionZip (CVPR’25), DSTP=true2026.04 | 39.22 | 79.7 | 81.04 | — | |
| DSTPBackbone=Qwen3-VL-4B, Token Retention Ratio=22.2%, Pruning Method=FastV (ECCV’24), DSTP=true2026.04 | 39.05 | 79.3 | 71.48 | — | |
| DSTPBackbone=Qwen3-VL-4B, Token Retention Ratio=22.2%, Pruning Method=VisionZip (CVPR’25), DSTP=true2026.04 | 37.58 | 76.4 | 68.74 | — | |
| DSTPBackbone=Qwen3-VL-4B, Token Retention Ratio=22.2%, Pruning Method=DivPrune (CVPR’25), DSTP=true2026.04 | 35.55 | 72.2 | 73.34 | — | |
| VisionZip (CVPR’25)Backbone=InternVL3.5-8B, Token Retention Ratio=33.3%, DSTP=false2026.04 | 34.45 | 66.1 | 59.54 | — | |
| DivPrune (CVPR’25)Backbone=InternVL3.5-8B, Token Retention Ratio=22.2%, DSTP=false2026.04 | 34 | 65.2 | 57.61 | — | |
| FastV (ECCV’24)Backbone=InternVL3.5-8B, Token Retention Ratio=33.3%, DSTP=false2026.04 | 33.08 | 63.5 | 56.66 | — | |
| DivPrune (CVPR’25)Backbone=InternVL3.5-8B, Token Retention Ratio=33.3%, DSTP=false2026.04 | 32.78 | 62.9 | 63.55 | — | |
| VisionZip (CVPR’25)Backbone=InternVL3.5-8B, Token Retention Ratio=22.2%, DSTP=false2026.04 | 31.76 | 60.9 | 51.04 | — | |
| FastV (ECCV’24)Backbone=Qwen3-VL-4B, Token Retention Ratio=33.3%, DSTP=false2026.04 | 30.64 | 62.3 | 55.68 | — | |
| DivPrune (CVPR’25)Backbone=Qwen3-VL-4B, Token Retention Ratio=33.3%, DSTP=false2026.04 | 30.64 | 62.3 | 55.24 | — | |
| FastV (ECCV’24)Backbone=InternVL3.5-8B, Token Retention Ratio=22.2%, DSTP=false2026.04 | 30.2 | 57.9 | 49.42 | — | |
| VisionZip (CVPR’25)Backbone=Qwen3-VL-4B, Token Retention Ratio=33.3%, DSTP=false2026.04 | 29.53 | 60 | 60.42 | — | |
| DivPrune (CVPR’25)Backbone=Qwen3-VL-4B, Token Retention Ratio=22.2%, DSTP=false2026.04 | 27.96 | 56.8 | 47.4 | — | |
| FastV (ECCV’24)Backbone=Qwen3-VL-4B, Token Retention Ratio=22.2%, DSTP=false2026.04 | 26.96 | 54.8 | 45.18 | — | |
| VisionZip (CVPR’25)Backbone=Qwen3-VL-4B, Token Retention Ratio=22.2%, DSTP=false2026.04 | 25.5 | 51.8 | 46.4 | — | |
| DRIFTModel Category=Reasoning Fine-tuning Methods2025.10 | — | — | — | 47.2 | |
| InternLM-XComposer2.5Model Category=Open-source Models2025.10 | — | — | — | 34.7 | |
| InternVL2-8BModel Category=Open-source Models2025.10 | — | — | — | 33.6 | |
| InternVL2.5-8BModel Category=Open-source Models2025.10 | — | — | — | 36 | |
| InternVL3-8BModel Category=Open-source Models2025.10 | — | — | — | 43.6 | |
| Kimi-VL-16BModel Category=Open-source Models2025.10 | — | — | — | 42.7 | |
| LLaVA-OneVision-7BModel Category=Open-source Models2025.10 | — | — | — | 32 | |
| OpenVLThinker-7BModel Category=Reasoning Fine-tuning Methods2025.10 | — | — | — | 44.5 | |
| QvQ-72B-PreviewModel Category=Open-source Models2025.10 | — | — | — | 58.2 | |
| Qwen2-VL-7BModel Category=Open-source Models2025.10 | — | — | — | 33.3 | |
| Qwen2.5-VL-7BModel Category=Open-source Models, Evaluation Source=Reproduced by authors2025.10 | — | — | — | 46.7 | |
| R1-Onevision-7BModel Category=Reasoning Fine-tuning Methods2025.10 | — | — | — | 61.8 | |
| SFT BaselineModel Category=Reasoning Fine-tuning Methods2025.10 | — | — | — | 45.6 |