Text-based Visual Question Answering on VQAText
86.1AccuracyAutoV
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| AutoVBackbone=Qwen2.5-VL 7B2025.06 | 86.1 | — | |
| APIBackbone=Qwen2.5-VL 7B2025.06 | 85.3 | — | |
| VanillaBackbone=Qwen2.5-VL-7B, Token Pruning Rate=0%2026.06 | 84.8 | — | |
| Qwen2.5-VL-7BRetain Tokens=1296, Retain Ratio (%)=100%2026.07 | 84.5 | — | |
| BaseBackbone=Qwen2.5-VL 7B2025.06 | 83 | — | |
| CircleBackbone=Qwen2.5-VL 7B2025.06 | 80.5 | — | |
| HoloVBackbone=Qwen2.5-VL-7B, Token Pruning Rate=66.7%2026.06 | 78.9 | — | |
| DivPruneRetain Tokens=512, Retain Ratio (%)=↓ 60.5%2026.07 | 78.8 | — | |
| EADPRetain Tokens=512, Retain Ratio (%)=↓ 60.5%2026.07 | 78.6 | — | |
| AutoVBackbone=LLaVA-OneVision 7B2025.06 | 78 | — | |
| AutoVBackbone=InternVL2 8B2025.06 | 78 | — | |
| FGVPBackbone=Qwen2.5-VL 7B2025.06 | 78 | — | |
| FastVBackbone=Qwen2.5-VL-7B, Token Pruning Rate=66.7%2026.06 | 77.9 | — | |
| APIBackbone=LLaVA-OneVision 7B2025.06 | 77.7 | — | |
| CircleBackbone=LLaVA-OneVision 7B2025.06 | 77.2 | — | |
| StepPruneBackbone=Qwen2.5-VL-7B, Token Pruning Rate=66.7%2026.06 | 77.2 | — | |
| BaseBackbone=InternVL2 8B2025.06 | 77 | — | |
| APIBackbone=InternVL2 8B2025.06 | 77 | — | |
| BaseBackbone=LLaVA-OneVision 7B2025.06 | 76.1 | — | |
| HiPruneRetain Tokens=512, Retain Ratio (%)=↓ 60.5%2026.07 | 75.8 | — | |
| StepPruneBackbone=Qwen2.5-VL-7B, Token Pruning Rate=77.8%2026.06 | 75.5 | — | |
| CircleBackbone=InternVL2 8B2025.06 | 74.9 | — | |
| DivPruneRetain Tokens=256, Retain Ratio (%)=↓ 80.2%2026.07 | 74 | — | |
| EADPRetain Tokens=256, Retain Ratio (%)=↓ 80.2%2026.07 | 73.8 | — | |
| StepPruneBackbone=Qwen2.5-VL-7B, Token Pruning Rate=88.9%2026.06 | 70.8 | — | |
| FGVPBackbone=InternVL2 8B2025.06 | 70.7 | — | |
| HoloVBackbone=Qwen2.5-VL-7B, Token Pruning Rate=77.8%2026.06 | 70.3 | — | |
| FastVBackbone=Qwen2.5-VL-7B, Token Pruning Rate=77.8%2026.06 | 69 | — | |
| CDPrunerRetain Tokens=512, Retain Ratio (%)=↓ 60.5%2026.07 | 66.2 | — | |
| DivPruneRetain Tokens=128, Retain Ratio (%)=↓ 90.1%2026.07 | 66.1 | — | |
| EADPRetain Tokens=128, Retain Ratio (%)=↓ 90.1%2026.07 | 65.8 | — | |
| VanillaToken Retention Ratio=100%, Backbone=LLaVA-NeXT-7B2026.02 | 65.5 | 100 | |
| FGVPBackbone=LLaVA-OneVision 7B2025.06 | 65.4 | — | |
| Upper BoundBase Model=LLaVA-NeXT-7B, Token Retention Budget=unpruned2026.04 | 64.9 | — | |
| HiPruneRetain Tokens=256, Retain Ratio (%)=↓ 80.2%2026.07 | 64.2 | — | |
| SwiftVLMToken Retention Ratio=33.3%, Backbone=LLaVA-NeXT-7B2026.02 | 64.1 | 98 | |
| AutoVBackbone=LLaVA-1.5 13B2025.06 | 62.7 | — | |
| FEATHERToken Retention Ratio=33.3%, Backbone=LLaVA-NeXT-7B2026.02 | 62.6 | 92.8 | |
| SwiftVLMToken Retention Ratio=22.2%, Backbone=LLaVA-NeXT-7B2026.02 | 62.4 | 97.1 | |
| APIBackbone=LLaVA-1.5 13B2025.06 | 62.1 | — | |
| CLASPBase Model=LLaVA-NeXT-7B, Token Retention Budget=640 Tokens2026.04 | 61.8 | — | |
| HoloVBackbone=Qwen2.5-VL-7B, Token Pruning Rate=88.9%2026.06 | 61.8 | — | |
| CLASPBase Model=LLaVA-NeXT-7B, Token Retention Budget=320 Tokens2026.04 | 61.7 | — | |
| LLaVA-NeXT-7BNumber of Tokens=2880, Token Reduction Ratio=0%2026.06 | 61.39 | — | |
| VanillaToken Budget=2880, Retention Ratio=100%2026.05 | 61.3 | 100 | |
| BaseBackbone=LLaVA-1.5 13B2025.06 | 61.2 | — | |
| FEATHERToken Retention Ratio=22.2%, Backbone=LLaVA-NeXT-7B2026.02 | 60.9 | 87.5 | |
| AutoVBackbone=LLaVA-1.5 7B2025.06 | 60.6 | — | |
| FastVBackbone=Qwen2.5-VL-7B, Token Pruning Rate=88.9%2026.06 | 60.3 | — | |
| VisionZipToken Budget=640, Retention Ratio=↓ 77.8%2026.05 | 60.2 | 97.4 | |
| SparseVLMBase Model=LLaVA-NeXT-7B, Token Retention Budget=640 Tokens2026.04 | 59.7 | — | |
| CLASPBase Model=LLaVA-NeXT-7B, Token Retention Budget=160 Tokens2026.04 | 59.3 | — | |
| LRCPToken Budget=640, Retention Ratio=↓ 77.8%2026.05 | 58.9 | 98.3 | |
| VisionZipToken Budget=320, Retention Ratio=↓ 88.9%2026.05 | 58.9 | 93.6 | |
| SparseVLMToken Budget=320, Retention Ratio=↓ 88.9%2026.05 | 58.8 | 93.7 | |
| FastVToken Retention Ratio=33.3%, Backbone=LLaVA-NeXT-7B2026.02 | 58.7 | 75.1 | |
| APIBackbone=LLaVA-1.5 7B2025.06 | 58.4 | — | |
| SparseVLMBase Model=LLaVA-NeXT-7B, Token Retention Budget=320 Tokens2026.04 | 58.4 | — | |
| ALVTSNumber of Tokens=320, Token Reduction Ratio=89%2026.06 | 58.37 | — | |
| BaseBackbone=LLaVA-1.5 7B2025.06 | 58.2 | — | |
| LLaVA-1.5-7BRetained Tokens=576, Pruning Ratio=0%, Base Model=LLaVA-1.5-7B2025.03 | 58.2 | 100 | |
| Upper BoundRetained Tokens=5762026.04 | 58.2 | — | |
| CircleBackbone=LLaVA-1.5 13B2025.06 | 58.1 | — | |
| TwigVLM++Retained Tokens=192, Pruning Ratio=66.7%, Base Model=LLaVA-1.5-7B2025.03 | 58 | 99.6 | |
| TwigVLM++Retained Tokens=128, Pruning Ratio=77.8%, Base Model=LLaVA-1.5-7B2025.03 | 58 | 99.2 | |
| ApETToken Budget=640, Retention Ratio=↓ 77.8%2026.05 | 57.9 | 97.6 | |
| SparseVLMToken Budget=640, Retention Ratio=↓ 77.8%2026.05 | 57.8 | 95.7 | |
| CLASPRetained Tokens=192, Pruning Ratio=66.7%2026.04 | 57.6 | — | |
| FastVNumber of Tokens=320, Token Reduction Ratio=89%2026.06 | 57.53 | — | |
| LRCPToken Budget=320, Retention Ratio=↓ 88.9%2026.05 | 57.4 | 95.3 | |
| DARTNumber of Tokens=320, Token Reduction Ratio=89%2026.06 | 57.02 | — | |
| PDropNumber of Tokens=320, Token Reduction Ratio=89%2026.06 | 56.93 | — | |
| TwigVLM++Retained Tokens=64, Pruning Ratio=88.9%, Base Model=LLaVA-1.5-7B2025.03 | 56.7 | 97.7 | |
| CLASPRetained Tokens=128, Pruning Ratio=77.8%2026.04 | 56.7 | — | |
| VisionZipToken Budget=160, Retention Ratio=↓ 94.4%2026.05 | 56.2 | 88.6 | |
| SparseVLMBase Model=LLaVA-NeXT-7B, Token Retention Budget=160 Tokens2026.04 | 55.9 | — | |
| LRCPToken Budget=160, Retention Ratio=↓ 94.4%2026.05 | 55.8 | 91.8 | |
| CircleBackbone=LLaVA-1.5 7B2025.06 | 55.5 | — | |
| CLASPRetained Tokens=64, Pruning Ratio=88.9%2026.04 | 55.2 | — | |
| CDPrunerRetain Tokens=256, Retain Ratio (%)=↓ 80.2%2026.07 | 55.1 | — | |
| ApETToken Budget=320, Retention Ratio=↓ 88.9%2026.05 | 54.4 | 94.2 | |
| ApETToken Budget=160, Retention Ratio=↓ 94.4%2026.05 | 53.8 | 90.1 | |
| FastVToken Retention Ratio=22.2%, Backbone=LLaVA-NeXT-7B2026.02 | 52.6 | 66.9 | |
| HiPruneRetain Tokens=128, Retain Ratio (%)=↓ 90.1%2026.07 | 51.1 | — | |
| FGVPBackbone=LLaVA-1.5 13B2025.06 | 49.2 | — | |
| SparseVLMToken Budget=160, Retention Ratio=↓ 94.4%2026.05 | 46.4 | 80.2 | |
| FGVPBackbone=LLaVA-1.5 7B2025.06 | 46.3 | — | |
| CDPrunerRetain Tokens=128, Retain Ratio (%)=↓ 90.1%2026.07 | 44.7 | — | |
| VTWNumber of Tokens=320, Token Reduction Ratio=89%2026.06 | 37.88 | — |