Multimodal Reasoning on MMBench v1.1 (test)
67.4AccuracyLLaVA
Evaluation results
| Method | Links | ||
|---|---|---|---|
| LLaVABackbone=LLaVA-NeXT-7B, Tokens Retained=2880, Retention Ratio=100%2025.08 | 67.4 | — | |
| HiPrune++Backbone=LLaVA-NeXT-7B, Tokens Retained=640, Retention Ratio=22.2%2025.08 | 67.2 | — | |
| HiPruneBackbone=LLaVA-NeXT-7B, Tokens Retained=640, Retention Ratio=22.2%2025.08 | 67 | — | |
| TRIMBackbone=LLaVA-NeXT-7B, Tokens Retained=640, Retention Ratio=22.2%2025.08 | 66.8 | — | |
| VisionZipBackbone=LLaVA-NeXT-7B, Tokens Retained=640, Retention Ratio=22.2%2025.08 | 66.3 | — | |
| HiPrune++Backbone=LLaVA-NeXT-7B, Tokens Retained=320, Retention Ratio=11.1%2025.08 | 66.2 | — | |
| HiREDBackbone=LLaVA-NeXT-7B, Tokens Retained=640, Retention Ratio=22.2%2025.08 | 66 | — | |
| SparseVLMBackbone=LLaVA-NeXT-7B, Tokens Retained=640, Retention Ratio=22.2%2025.08 | 65.9 | — | |
| HiPruneBackbone=LLaVA-NeXT-7B, Tokens Retained=320, Retention Ratio=11.1%2025.08 | 65.3 | — | |
| VisPrunerBackbone=LLaVA-NeXT-7B, Tokens Retained=640, Retention Ratio=22.2%2025.08 | 65.2 | — | |
| DivPruneBackbone=LLaVA-NeXT-7B, Tokens Retained=640, Retention Ratio=22.2%2025.08 | 65 | — | |
| HiREDBackbone=LLaVA-NeXT-7B, Tokens Retained=320, Retention Ratio=11.1%2025.08 | 64.2 | — | |
| PDropBackbone=LLaVA-NeXT-7B, Tokens Retained=640, Retention Ratio=22.2%2025.08 | 64.1 | — | |
| DivPruneBackbone=LLaVA-NeXT-7B, Tokens Retained=320, Retention Ratio=11.1%2025.08 | 63.9 | — | |
| VisPrunerBackbone=LLaVA-NeXT-7B, Tokens Retained=320, Retention Ratio=11.1%2025.08 | 63.8 | — | |
| TRIMBackbone=LLaVA-NeXT-7B, Tokens Retained=320, Retention Ratio=11.1%2025.08 | 63.5 | — | |
| FastVBackbone=LLaVA-NeXT-7B, Tokens Retained=640, Retention Ratio=22.2%2025.08 | 63.1 | — | |
| VisionZipBackbone=LLaVA-NeXT-7B, Tokens Retained=320, Retention Ratio=11.1%2025.08 | 63.1 | — | |
| SparseVLMBackbone=LLaVA-NeXT-7B, Tokens Retained=320, Retention Ratio=11.1%2025.08 | 63.1 | — | |
| DivPruneBackbone=LLaVA-NeXT-7B, Tokens Retained=160, Retention Ratio=5.6%2025.08 | 62.5 | — | |
| TRIMBackbone=LLaVA-NeXT-7B, Tokens Retained=160, Retention Ratio=5.6%2025.08 | 61.6 | — | |
| HiPrune++Backbone=LLaVA-NeXT-7B, Tokens Retained=160, Retention Ratio=5.6%2025.08 | 61.5 | — | |
| VisionZipBackbone=LLaVA-NeXT-7B, Tokens Retained=160, Retention Ratio=5.6%2025.08 | 60.1 | — | |
| HiPruneBackbone=LLaVA-NeXT-7B, Tokens Retained=160, Retention Ratio=5.6%2025.08 | 59.8 | — | |
| VisPrunerBackbone=LLaVA-NeXT-7B, Tokens Retained=160, Retention Ratio=5.6%2025.08 | 59.2 | — | |
| PDropBackbone=LLaVA-NeXT-7B, Tokens Retained=320, Retention Ratio=11.1%2025.08 | 55.5 | — | |
| FastVBackbone=LLaVA-NeXT-7B, Tokens Retained=320, Retention Ratio=11.1%2025.08 | 53.4 | — | |
| CLASPBackbone=Qwen2.5-VL-7B, Token Pruning Rate=66.7%2026.04 | — | 78.6 | |
| CLASPBackbone=Qwen2.5-VL-7B, Token Pruning Rate=77.8%2026.04 | — | 76.2 | |
| CLASPBackbone=Qwen2.5-VL-7B, Token Pruning Rate=88.9%2026.04 | — | 72.7 | |
| DARTToken Budget=Retain 33.30% Tokens, Venue=EMNLP'25, Backbone=Qwen-2.5-VL-7B2026.02 | — | 0.809 | |
| DARTToken Budget=Retain 22.2% Tokens, Venue=EMNLP'25, Backbone=Qwen-2.5-VL-7B2026.02 | — | 0.789 | |
| DARTToken Budget=Retain 11.1% Tokens, Venue=EMNLP'25, Backbone=Qwen-2.5-VL-7B2026.02 | — | 0.733 | |
| PIO-FVLMToken Budget=Retain 33.30% Tokens, Venue=Ours, Backbone=Qwen-2.5-VL-7B2026.02 | — | 0.821 | |
| PIO-FVLMToken Budget=Retain 22.2% Tokens, Venue=Ours, Backbone=Qwen-2.5-VL-7B2026.02 | — | 0.809 | |
| PIO-FVLMToken Budget=Retain 11.1% Tokens, Venue=Ours, Backbone=Qwen-2.5-VL-7B2026.02 | — | 0.776 | |
| SparseVLMBackbone=Qwen2.5-VL-7B, Token Pruning Rate=66.7%2026.04 | — | 78.4 | |
| SparseVLMBackbone=Qwen2.5-VL-7B, Token Pruning Rate=77.8%2026.04 | — | 75.6 | |
| SparseVLMBackbone=Qwen2.5-VL-7B, Token Pruning Rate=88.9%2026.04 | — | 72.1 | |
| Upper BoundBackbone=Qwen2.5-VL-7B2026.04 | — | 82.2 | |
| VanillaToken Budget=1296 Tokens (Upper Bound), Backbone=Qwen-2.5-VL-7B2026.02 | — | 0.837 |