Visual Question Answering on TextVQA (Accuracy, Relative Accuracy)
97AccuracyOracle
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| OracleStrategy=Oracle2026.05 | 97 | — | |
| Cloud-OnlyStrategy=Cloud-Only2026.05 | 81.9 | — | |
| VanillaModel=Qwen3-VL-4B, Token Retention Ratio=100%2026.04 | 81.57 | — | |
| FastV w/ DSTPModel=Qwen3-VL-4B, Token Retention Ratio=33.3%2026.04 | 77.95 | 95.6 | |
| VanillaModel=InternVL3.5-8B, Token Retention Ratio=100%2026.04 | 77.59 | — | |
| INAR-VLStrategy=INAR-VL2026.05 | 77.4 | — | |
| Static routingStrategy=Static routing2026.05 | 77.4 | — | |
| VisionZip w/ DSTPModel=Qwen3-VL-4B, Token Retention Ratio=33.3%2026.04 | 77.2 | 94.6 | |
| VisionZipModel=Qwen3-VL-4B, Token Retention Ratio=33.3%2026.04 | 77.1 | 94.5 | |
| Text-OnlyStrategy=Text-Only2026.05 | 76.5 | — | |
| FastVModel=Qwen3-VL-4B, Token Retention Ratio=33.3%2026.04 | 74.82 | 91.7 | |
| FastV w/ DSTPModel=Qwen3-VL-4B, Token Retention Ratio=22.2%2026.04 | 74.47 | 91.3 | |
| DivPrune w/ DSTPModel=Qwen3-VL-4B, Token Retention Ratio=33.3%2026.04 | 74.11 | 90.9 | |
| DivPrune w/ DSTPModel=InternVL3.5-8B, Token Retention Ratio=33.3%2026.04 | 73.75 | 95.1 | |
| DivPrune w/ DSTPModel=Qwen3-VL-4B, Token Retention Ratio=22.2%2026.04 | 72.87 | 89.3 | |
| Image-OnlyStrategy=Image-Only2026.05 | 72.4 | — | |
| FastVModel=Qwen3-VL-4B, Token Retention Ratio=22.2%2026.04 | 72.35 | 88.7 | |
| VisionZip w/ DSTPModel=Qwen3-VL-4B, Token Retention Ratio=22.2%2026.04 | 71.97 | 88.2 | |
| VisionZipModel=Qwen3-VL-4B, Token Retention Ratio=22.2%2026.04 | 71.81 | 88 | |
| DivPruneModel=Qwen3-VL-4B, Token Retention Ratio=33.3%2026.04 | 71.36 | 87.5 | |
| DivPruneModel=Qwen3-VL-4B, Token Retention Ratio=22.2%2026.04 | 70.04 | 85.9 | |
| DivPrune w/ DSTPModel=InternVL3.5-8B, Token Retention Ratio=22.2%2026.04 | 69.39 | 89.4 | |
| Edge-OnlyStrategy=Edge-Only2026.05 | 69 | — | |
| FastV w/ DSTPModel=InternVL3.5-8B, Token Retention Ratio=33.3%2026.04 | 67.74 | 87.3 | |
| VisionZip w/ DSTPModel=InternVL3.5-8B, Token Retention Ratio=33.3%2026.04 | 66.29 | 85.4 | |
| FastV w/ DSTPModel=InternVL3.5-8B, Token Retention Ratio=22.2%2026.04 | 66.03 | 85.1 | |
| FastVModel=InternVL3.5-8B, Token Retention Ratio=33.3%2026.04 | 64.92 | 83.7 | |
| FastVModel=InternVL3.5-8B, Token Retention Ratio=22.2%2026.04 | 63.18 | 81.4 | |
| VisionZip w/ DSTPModel=InternVL3.5-8B, Token Retention Ratio=22.2%2026.04 | 62.06 | 80 | |
| DivPruneModel=InternVL3.5-8B, Token Retention Ratio=33.3%2026.04 | 61.88 | 79.8 | |
| VisionZipModel=InternVL3.5-8B, Token Retention Ratio=33.3%2026.04 | 61.21 | 78.9 | |
| MolmoE-1B-7BNA=1.58B, Avgk=8.002026.06 | 56.37 | — | |
| Individuals2026.05 | 55.89 | — | |
| DiDi-Merging2026.05 | 55.88 | — | |
| FREE-Merging2026.05 | 55.84 | — | |
| DivPruneModel=InternVL3.5-8B, Token Retention Ratio=22.2%2026.04 | 55.77 | 71.9 | |
| MMER2026.05 | 55.74 | — | |
| DYNMoENA=1.40B, Avgk=6.192026.06 | 55.16 | — | |
| GeMoENA=1.32B, Avgk=5.432026.06 | 54.88 | — | |
| AdaMoENA=1.39B, Avgk=6.132026.06 | 52.31 | — | |
| Top-pNA=2.05B, Avgk=12.652026.06 | 51.64 | — | |
| VisionZipModel=InternVL3.5-8B, Token Retention Ratio=22.2%2026.04 | 51.04 | 65.8 | |
| MoE++NA=1.48B, Avgk=7.032026.06 | 46.04 | — | |
| NaiveMC2026.05 | 42.29 | — |