Multimodal Understanding on LLaVA Evaluation Suite 1.5
101.15Average ScoreMS-Resampler
Evaluation Results
| Method | Links | ||||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| MS-ResamplerBackbone=LLaVA-1.5-7B, Vision token count=1442026.06 | 101.15 | 62 | 65.6 | — | 87 | 71.2 | 78 | — | — | 33.1 | — | — | — | 65.2 | — | — | — | — | — | 57.2 | 1,453 | 33.9 | |
| TokenPackerBackbone=LLaVA-1.5-7B, Vision token count=1442026.06 | 100.25 | 62 | 65 | — | 87 | 68.9 | 78 | — | — | 33 | — | — | — | 65.1 | — | — | — | — | — | 57.1 | 1,454 | 32.6 | |
| MS-ResamplerBackbone=LLaVA-1.5-7B, Vision token count=642026.06 | 100.07 | 62.1 | 64.5 | — | 87.6 | 70.6 | 78 | — | — | 30.8 | — | — | — | 66.1 | — | — | — | — | — | 56.6 | 1,469 | 32.9 | |
| VanillaToken Count=576, Reduction Ratio=0%2026.01 | 100 | 61.9 | 64.7 | 1,862 | 85.9 | 69.5 | 78.5 | 58.2 | 58.6 | 31.1 | 50.1 | — | — | — | — | — | — | — | — | — | — | — | |
| LLaVA-1.5-7B# Vision Tokens=5762026.02 | 100 | 61.9 | — | 1,864 | 85.9 | 69.5 | 78.5 | 58.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LLaVA-1.5-13BRetained Tokens=576, Pruning Rate=0.0%2025.09 | 100 | 63.2 | 67.7 | 1,818 | 85.9 | 72.8 | 80 | 61.3 | 66.9 | — | — | — | 36.4 | — | — | — | — | — | — | — | — | — | |
| LLaVA-1.5 Vanilla 13BTokens Retained=576, Compression Ratio=100%2025.08 | 100 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VanillaLoc.=-, Retained Tokens=576, Pruning Ratio=0%2026.03 | 100 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VanillaEarly Cmp.=−, Retained Tokens=576, Compression Ratio=100%2026.03 | 100 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VanillaBase Model=LLaVA-1.5 13B, Retained Tokens=576 (100%)2026.03 | 100 | 63.2 | 67.7 | 1,818 | 85.9 | 72.8 | 80 | 61.3 | — | — | 56.6 | — | 36.4 | 66.9 | — | — | — | — | — | — | — | — | |
| VanillaBackbone=LLaVA-1.5-7B, Vision token count=1442026.06 | 100 | 62 | 64.3 | — | 85.9 | 68 | 78.5 | — | — | 31.6 | — | — | — | 66.1 | — | — | — | — | — | 58.2 | 1,511 | 31.9 | |
| VanillaBackbone=LLaVA-1.5-13B, Vision token count=1442026.06 | 100 | 63.3 | 67.7 | — | 87.9 | 71.6 | 80 | — | — | 35.4 | — | — | — | 68.2 | — | — | — | — | — | 63.8 | 1,531 | 33.1 | |
| MS-ResamplerBackbone=LLaVA-1.5-7B, Vision token count=362026.06 | 99.32 | 61.2 | 64.9 | — | 87.2 | 69.9 | 76.8 | — | — | 32.8 | — | — | — | 62.8 | — | — | — | — | — | 55.8 | 1,429 | 32.6 | |
| SAEPBackbone=LLaVA-1.5-7B, Vision token count=1442026.06 | 99.31 | 61.8 | 63.8 | — | 87.2 | 69 | 77.6 | — | — | 32.8 | — | — | — | 64.8 | — | — | — | — | — | 56.4 | 1,424 | 31.7 | |
| LDP-v2Backbone=LLaVA-1.5-7B, Vision token count=1442026.06 | 99.25 | 61.5 | 65.1 | — | 86.3 | 70.9 | 77.3 | — | — | 31.7 | — | — | — | 63.9 | — | — | — | — | — | 56.4 | 1,435 | 31.9 | |
| MS-ResamplerBackbone=LLaVA-1.5-13B, Vision token count=1442026.06 | 98.77 | 62.4 | 66.9 | — | 87.9 | 72.8 | 78.3 | — | — | 35.1 | — | — | — | 66.2 | — | — | — | — | — | 58.5 | 1,547 | 33.7 | |
| VisionZip 🔥Tokens Retained=192, Compression Ratio=↓ 66.7%2025.08 | 98.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MMTokTokens Retained=192, Compression Ratio=↓ 66.7%2025.08 | 98.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LDP-v2Backbone=LLaVA-1.5-7B, Vision token count=642026.06 | 98.61 | 60.1 | 66.2 | — | 86.8 | 71.4 | 76.1 | — | — | 30.1 | — | — | — | 62.6 | — | — | — | — | — | 55.1 | 1,414 | 33.5 | |
| ZOO-PruneRetained Tokens=192, Pruning Rate=66.7%2025.09 | 98.6 | 59.95 | 66.67 | 1,762.41 | 86.73 | 73.12 | 78.65 | 59.11 | 65.56 | — | — | — | 37.33 | — | — | — | — | — | — | — | — | — | |
| PRUNESID-DynEarly Cmp.=✓, Retained Tokens=192, Compression Ratio=66.7%2026.03 | 98.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PRUNESIDLoc.=Pre, Retained Tokens=192, Pruning Ratio=66.7%2026.03 | 98.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TokenPackerBackbone=LLaVA-1.5-13B, Vision token count=1442026.06 | 98.22 | 62.3 | 67.4 | — | 87.4 | 70.7 | 78.8 | — | — | 34.5 | — | — | — | 65.4 | — | — | — | — | — | 58.8 | 1,525 | 33.9 | |
| ADSC# Vision Tokens=1922026.02 | 98.2 | 61 | — | 1,801 | 85 | 69.5 | 77.1 | 56.2 | — | — | — | 1.8 | — | — | — | — | — | — | — | — | — | — | |
| DivPruneTokens Retained=192, Compression Ratio=↓ 66.7%2025.08 | 98.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| HoloVLoc.=Pre, Retained Tokens=192, Pruning Ratio=66.7%2026.03 | 98.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| AutoSelectLoc.=Pre, Retained Tokens=192, Pruning Ratio=66.7%2026.03 | 98.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DivPruneRetained Tokens=192, Pruning Rate=66.7%2025.09 | 98.1 | 59.42 | 66.58 | 1,781.5 | 86.76 | 72.88 | 77.98 | 58.46 | 65.72 | — | — | — | 36.56 | — | — | — | — | — | — | — | — | — | |
| PRUNESID-DynEarly Cmp.=✓, Retained Tokens=128, Compression Ratio=77.8%2026.03 | 98.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LDP-v2Backbone=LLaVA-1.5-13B, Vision token count=1442026.06 | 98.03 | 62 | 66.9 | — | 87.6 | 72.3 | 78.4 | — | — | 34.4 | — | — | — | 65.3 | — | — | — | — | — | 58.4 | 1,533 | 33.2 | |
| PRUNESIDBase Model=LLaVA-1.5 13B, Retained Tokens=192 (↓ 66.7%)2026.03 | 98 | 59.6 | 65.9 | 1,770 | 86.4 | 72.8 | 78 | 58.6 | — | — | 56 | — | 36.9 | 65.2 | — | — | — | — | — | — | — | — | |
| VisionZipRetained Tokens=192, Pruning Rate=66.7%2025.09 | 97.9 | 59.1 | 66.9 | 1,754 | 85.1 | 73.5 | 78.1 | 59.5 | 65.2 | — | — | — | 36.4 | — | — | — | — | — | — | — | — | — | |
| VisionZipTokens Retained=192, Compression Ratio=↓ 66.7%2025.08 | 97.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VisionZipLoc.=Pre, Retained Tokens=192, Pruning Ratio=66.7%2026.03 | 97.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ZOO-PruneRetained Tokens=128, Pruning Rate=77.8%2025.09 | 97.8 | 58.89 | 67.01 | 1,791.1 | 86.95 | 73.38 | 77.83 | 58.8 | 64.5 | — | — | — | 35.56 | — | — | — | — | — | — | — | — | — | |
| DivPruneLoc.=Pre, Retained Tokens=192, Pruning Ratio=66.7%2026.03 | 97.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VisionZipBase Model=LLaVA-1.5 13B, Retained Tokens=192 (↓ 66.7%)2026.03 | 97.8 | 59.6 | 65.9 | 1,770 | 86.4 | 72.8 | 78 | 58.6 | — | — | 54.9 | — | 36.9 | 65.2 | — | — | — | — | — | — | — | — | |
| LDP-v2Backbone=LLaVA-1.5-7B, Vision token count=362026.06 | 97.77 | 59.9 | 64.8 | — | 86.4 | 69.7 | 75.2 | — | — | 31.7 | — | — | — | 61 | — | — | — | — | — | 53.6 | 1,443 | 32.3 | |
| AutoSelectLoc.=Pre, Retained Tokens=128, Pruning Ratio=77.8%2026.03 | 97.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TokenPackerBackbone=LLaVA-1.5-7B, Vision token count=642026.06 | 97.59 | 61 | 64.4 | — | 86.5 | 69.4 | 77.1 | — | — | 29.1 | — | — | — | 63 | — | — | — | — | — | 55.6 | 1,440 | 31.3 | |
| FasterVLMBackbone=LLaVA-1.5-7B, Vision token count=1442026.06 | 97.54 | 58 | 64.3 | — | 83.5 | 67.8 | 76.2 | — | — | 32.2 | — | — | — | 62.3 | — | — | — | — | — | 57.2 | 1,440 | 31.3 | |
| MMTokTokens Retained=128, Compression Ratio=↓ 77.8%2025.08 | 97.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ADSC# Vision Tokens=1282026.02 | 97.4 | 60.1 | — | 1,782 | 84.5 | 69.9 | 76.3 | 55.1 | — | — | — | 2.6 | — | — | — | — | — | — | — | — | — | — | |
| VisionZip 🔥Tokens Retained=128, Compression Ratio=↓ 77.8%2025.08 | 97.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PRUNESIDBase Model=LLaVA-1.5 13B, Retained Tokens=128 (↓ 77.8%)2026.03 | 97.4 | 58.9 | 65.5 | 1,811 | 85.9 | 73.1 | 76.7 | 57.5 | — | — | 56.8 | — | 35.8 | 64.1 | — | — | — | — | — | — | — | — | |
| TokenPackerBackbone=LLaVA-1.5-7B, Vision token count=362026.06 | 97.39 | 60.5 | 63.5 | — | 86.5 | 67.8 | 76.1 | — | — | 30.2 | — | — | — | 62.5 | — | — | — | — | — | 55.2 | 1,420 | 32.3 | |
| DARTLoc.=L2, Retained Tokens=192, Pruning Ratio=66.7%2026.03 | 97.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SAEPBackbone=LLaVA-1.5-7B, Vision token count=642026.06 | 97.16 | 60.3 | 62.9 | — | 86 | 68 | 75.6 | — | — | 30.5 | — | — | — | 63.4 | — | — | — | — | — | 53.9 | 1,380 | 33.1 | |
| MQT-LLaVABackbone=LLaVA-1.5-7B, Vision token count=1442026.06 | 97.09 | 61.4 | 64.4 | — | 86.2 | 67.6 | 76.8 | — | — | 29.8 | — | — | — | 62 | — | — | — | — | — | 52.6 | 1,445 | 31.9 | |
| VisionZipRetained Tokens=128, Pruning Rate=77.8%2025.09 | 97 | 57.9 | 66.7 | 1,743 | 85.2 | 74 | 76.8 | 58.7 | 63.8 | — | — | — | 36.1 | — | — | — | — | — | — | — | — | — | |
| DivPruneRetained Tokens=128, Pruning Rate=77.8%2025.09 | 97 | 58.89 | 66.07 | 1,748.56 | 86.53 | 72.83 | 77.1 | 58.17 | 64.22 | — | — | — | 35.56 | — | — | — | — | — | — | — | — | — | |
| VisionZipTokens Retained=128, Compression Ratio=↓ 77.8%2025.08 | 97 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VisionZipBase Model=LLaVA-1.5 13B, Retained Tokens=128 (↓ 77.8%)2026.03 | 97 | 57.9 | 66.7 | 1,743 | 85.2 | 74 | 76.8 | 58.7 | — | — | 55 | — | 36.1 | 63.8 | — | — | — | — | — | — | — | — | |
| Pixel-ShuffleBackbone=LLaVA-1.5-7B, Vision token count=1442026.06 | 96.92 | 60.2 | 64.5 | — | 87 | 69.6 | 75.6 | — | — | 29.1 | — | — | — | 61.6 | — | — | — | — | — | 54.2 | 1,422 | 31.7 | |
| DivPruneTokens Retained=128, Compression Ratio=↓ 77.8%2025.08 | 96.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DivPruneLoc.=Pre, Retained Tokens=128, Pruning Ratio=77.8%2026.03 | 96.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PRUNESIDLoc.=Pre, Retained Tokens=128, Pruning Ratio=77.8%2026.03 | 96.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| HoloVLoc.=Pre, Retained Tokens=128, Pruning Ratio=77.8%2026.03 | 96.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VisionZipLoc.=Pre, Retained Tokens=128, Pruning Ratio=77.8%2026.03 | 96.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ZOO-PruneRetained Tokens=64, Pruning Rate=88.9%2025.09 | 96.5 | 58.58 | 64.78 | 1,780.03 | 85.34 | 72.09 | 76.39 | 58.59 | 63.02 | — | — | — | 36 | — | — | — | — | — | — | — | — | — | |
| AutoSelectLoc.=Pre, Retained Tokens=64, Pruning Ratio=88.9%2026.03 | 96.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MMTokTokens Retained=64, Compression Ratio=↓ 88.9%2025.08 | 96.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PRUNESID-DynEarly Cmp.=✓, Retained Tokens=64, Compression Ratio=88.9%2026.03 | 96.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PyramidDrop# Vision Tokens=1922026.02 | 96.2 | 57.3 | — | 1,797 | 82.3 | 69.2 | 75.1 | 56.5 | — | — | — | 3.8 | — | — | — | — | — | — | — | — | — | — | |
| DARTLoc.=L2, Retained Tokens=128, Pruning Ratio=77.8%2026.03 | 96.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FasterVLMBackbone=LLaVA-1.5-13B, Vision token count=1442026.06 | 96.17 | 59.1 | 66.4 | — | 86.1 | 71.9 | 77.1 | — | — | 33.6 | — | — | — | 63.8 | — | — | — | — | — | 58.8 | 1,489 | 32.3 | |
| PDropLoc.=L8–L24, Retained Tokens=192, Pruning Ratio=66.7%2026.03 | 96 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Pixel-ShuffleBackbone=LLaVA-1.5-13B, Vision token count=1442026.06 | 95.96 | 61.2 | 66.2 | — | 86.9 | 71.3 | 76.9 | — | — | 33.1 | — | — | — | 64 | — | — | — | — | — | 57 | 1,483 | 32.1 | |
| C-AbstractorBackbone=LLaVA-1.5-7B, Vision token count=1442026.06 | 95.91 | 60.4 | 63.3 | — | 85.9 | 68.5 | 75.7 | — | — | 28.2 | — | — | — | 62.3 | — | — | — | — | — | 54.1 | 1,392 | 31.1 | |
| SparseVLMLoc.=L1–L32, Retained Tokens=192, Pruning Ratio=66.7%2026.03 | 95.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DivPruneRetained Tokens=64, Pruning Rate=88.9%2025.09 | 95.4 | 57.66 | 64.6 | 1,777.93 | 84.8 | 71.34 | 75.2 | 57.11 | 62.44 | — | — | — | 35.22 | — | — | — | — | — | — | — | — | — | |
| DivPruneTokens Retained=64, Compression Ratio=↓ 88.9%2025.08 | 95.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SAEPBackbone=LLaVA-1.5-7B, Vision token count=362026.06 | 95.3 | 59.3 | 60.5 | — | 85.4 | 67.6 | 76.3 | — | — | 30.9 | — | — | — | 61.7 | — | — | — | — | — | 51.7 | 1,356 | 31.1 | |
| MQT-LLaVABackbone=LLaVA-1.5-7B, Vision token count=642026.06 | 95.25 | 60 | 63.5 | — | 83.6 | 67 | 75.3 | — | — | 28.9 | — | — | — | 60.8 | — | — | — | — | — | 51.6 | 1,454 | 30.9 | |
| PyramidDrop# Vision Tokens=1282026.02 | 95.2 | 57.1 | — | 1,761 | 82.3 | 68.4 | 72.9 | 56.6 | — | — | — | 4.8 | — | — | — | — | — | — | — | — | — | — | |
| PRUNESIDBase Model=LLaVA-1.5 13B, Retained Tokens=64 (↓ 88.9%)2026.03 | 95.2 | 57.8 | 63.8 | 1,711 | 82 | 71.8 | 75.2 | 56.3 | — | — | 57.3 | — | 35.7 | 62.8 | — | — | — | — | — | — | — | — | |
| ADSC# Vision Tokens=642026.02 | 95.1 | 58.4 | — | 1,701 | 84 | 69.8 | 73.9 | 53.2 | — | — | — | 4.9 | — | — | — | — | — | — | — | — | — | — | |
| PRUNESIDLoc.=Pre, Retained Tokens=64, Pruning Ratio=88.9%2026.03 | 95.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DivPruneLoc.=Pre, Retained Tokens=64, Pruning Ratio=88.9%2026.03 | 94.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VisionZip 🔥Tokens Retained=64, Compression Ratio=↓ 88.9%2025.08 | 94.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| HoloVLoc.=Pre, Retained Tokens=64, Pruning Ratio=88.9%2026.03 | 94.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MQT-LLaVABackbone=LLaVA-1.5-7B, Vision token count=362026.06 | 94.73 | 60 | 63.4 | — | 81.9 | 66.8 | 75.3 | — | — | 27.8 | — | — | — | 59 | — | — | — | — | — | 51 | 1,464 | 32.1 | |
| PDropLoc.=L8–L24, Retained Tokens=128, Pruning Ratio=77.8%2026.03 | 94.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VisionTrim ‡Token Count=16, Reduction Ratio=97.2%, Variant=Text-Guided Vision Complement2026.01 | 94.3 | 55 | 61.3 | 1,612 | 78.1 | 70.2 | 72.3 | 55.4 | 55.9 | 30.2 | 50.5 | — | — | — | — | — | — | — | — | — | — | — | |
| C-AbstractorBackbone=LLaVA-1.5-7B, Vision token count=642026.06 | 94.24 | 59.1 | 61.8 | — | 85.7 | 68.6 | 73.7 | — | — | 26.1 | — | — | — | 60 | — | — | — | — | — | 51.3 | 1,415 | 32.3 | |
| VisionZipBase Model=LLaVA-1.5 13B, Retained Tokens=64 (↓ 88.9%)2026.03 | 94.2 | 56.2 | 64.9 | 1,676 | 76 | 74.4 | 73.7 | 57.4 | — | — | 55.9 | — | 36.4 | 60.4 | — | — | — | — | — | — | — | — | |
| Pixel-ShuffleBackbone=LLaVA-1.5-7B, Vision token count=642026.06 | 94.18 | 58.7 | 64.4 | — | 85.5 | 68.7 | 73.2 | — | — | 25.8 | — | — | — | 59.3 | — | — | — | — | — | 52.2 | 1,383 | 32.1 | |
| C-AbstractorBackbone=LLaVA-1.5-7B, Vision token count=362026.06 | 93.97 | 58.3 | 62.4 | — | 85.7 | 69 | 72.4 | — | — | 29.3 | — | — | — | 58.4 | — | — | — | — | — | 49.6 | 1,347 | 31.8 | |
| VisionZipRetained Tokens=64, Pruning Rate=88.9%2025.09 | 93.7 | 56.2 | 64.9 | 1,676 | 76 | 74.4 | 73.7 | 57.4 | 60.4 | — | — | — | 36.4 | — | — | — | — | — | — | — | — | — | |
| VisionZipTokens Retained=64, Compression Ratio=↓ 88.9%2025.08 | 93.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| HiREDLoc.=Pre, Retained Tokens=192, Pruning Ratio=66.7%2026.03 | 93.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FasterVLMBackbone=LLaVA-1.5-7B, Vision token count=642026.06 | 93.48 | 55 | 61.6 | — | 80.4 | 68.8 | 72.5 | — | — | 28.7 | — | — | — | 57.9 | — | — | — | — | — | 56 | 1,342 | 31.8 | |
| VisionZipLoc.=Pre, Retained Tokens=64, Pruning Ratio=88.9%2026.03 | 92.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SparseVLMLoc.=L1–L32, Retained Tokens=128, Pruning Ratio=77.8%2026.03 | 92.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DARTLoc.=L2, Retained Tokens=64, Pruning Ratio=88.9%2026.03 | 92.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| HiREDLoc.=Pre, Retained Tokens=128, Pruning Ratio=77.8%2026.03 | 92.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VisionTrim ‡Token Count=8, Reduction Ratio=98.6%, Variant=Text-Guided Vision Complement2026.01 | 91.8 | 51.7 | 58.3 | 1,525 | 75.2 | 69.9 | 70.2 | 55.3 | 54.4 | 30.4 | 49.9 | — | — | — | — | — | — | — | — | — | — | — | |
| C-AbstractorBackbone=LLaVA-1.5-13B, Vision token count=1442026.06 | 91.75 | 60.4 | 63 | — | 86.3 | 70.4 | 74.8 | — | — | 28.2 | — | — | — | 61 | — | — | — | — | — | 53.9 | 1,391 | 31.3 | |
| Pixel-ShuffleBackbone=LLaVA-1.5-7B, Vision token count=362026.06 | 91.6 | 57.2 | 61.3 | — | 84.6 | 68.7 | 70.6 | — | — | 25.8 | — | — | — | 56.1 | — | — | — | — | — | 48.1 | 1,309 | 32.9 | |
| VisionTrimToken Count=16, Reduction Ratio=97.2%, Variant=Standard2026.01 | 90 | 51.5 | 57.7 | 1,510 | 72.6 | 69.4 | 68.4 | 53 | 53.4 | 28.9 | 50 | — | — | — | — | — | — | — | — | — | — | — |