Text-based Visual Question Answering on TextVQA
88.69TextVQA AccuracyBF16
Evaluation Results
| Method | Links | |
|---|---|---|
| BF16Model=Kimi-VL-A3B-Instruct, Avg-Bit=BF162026.06 | 88.69 | |
| BAGEL (Ours)Baseline=BAGEL, Params=7B active2026.06 | 88.5 | |
| MODEModel=Kimi-VL-A3B-Instruct, Avg-Bit=W32026.06 | 86.24 | |
| BAGELBaseline=BAGEL, Params=7B active2026.06 | 86 | |
| BLIP3o-8B (Ours)Baseline=BLIP3o-8B, Params=8B2026.06 | 85.2 | |
| MC-MoEModel=Kimi-VL-A3B-Instruct, Avg-Bit=W32026.06 | 85.06 | |
| VARGPT-v1.1 (Ours)Baseline=VARGPT-v1.1, Params=7B+2B2026.06 | 84.8 | |
| VEQ-MEModel=Kimi-VL-A3B-Instruct, Avg-Bit=W32026.06 | 84.24 | |
| BF16Model=Qwen3-VL-30B-A3B-Instruct, Avg-Bit=BF162026.06 | 83.37 | |
| BLIP3o-8BBaseline=BLIP3o-8B, Params=8B2026.06 | 83.1 | |
| Speed-QModel=Kimi-VL-A3B-Instruct, Avg-Bit=W32026.06 | 83.01 | |
| MoEQuantModel=Kimi-VL-A3B-Instruct, Avg-Bit=W32026.06 | 82.93 | |
| GPTQModel=Kimi-VL-A3B-Instruct, Avg-Bit=W32026.06 | 82.05 | |
| VARGPT-v1.1Baseline=VARGPT-v1.1, Params=7B+2B2026.06 | 82 | |
| MBQModel=Kimi-VL-A3B-Instruct, Avg-Bit=W32026.06 | 81.73 | |
| VanillaAverage Token Reduction=0.0%, Backbone=Qwen3-VL-8B2026.06 | 81.6 | |
| MODEModel=Qwen3-VL-30B-A3B-Instruct, Avg-Bit=W32026.06 | 80.65 | |
| Qwen2-VL-2B#Params=2B, TFLOPs=0.42025.06 | 79.7 | |
| LaTtE-Flow#Params=2B, TFLOPs=0.42025.06 | 79.7 | |
| MODEModel=Kimi-VL-A3B-Instruct, Avg-Bit=W22026.06 | 79.38 | |
| VEQ-MEModel=Qwen3-VL-30B-A3B-Instruct, Avg-Bit=W32026.06 | 78.58 | |
| Speed-QModel=Qwen3-VL-30B-A3B-Instruct, Avg-Bit=W32026.06 | 78.25 | |
| MBQModel=Qwen3-VL-30B-A3B-Instruct, Avg-Bit=W32026.06 | 77.98 | |
| MC-MoEModel=Qwen3-VL-30B-A3B-Instruct, Avg-Bit=W32026.06 | 77.93 | |
| PDrop+RerouteAverage Token Reduction=66.7%, Backbone=Qwen3-VL-8B2026.06 | 77.8 | |
| MoEQuantModel=Qwen3-VL-30B-A3B-Instruct, Avg-Bit=W32026.06 | 77.05 | |
| VanillaBackbone=InternVL-2.5-8B, Token Reduction Ratio=0%2026.06 | 76.8 | |
| FastVAverage Token Reduction=66.7%, Backbone=Qwen3-VL-8B2026.06 | 76.7 | |
| GPTQModel=Qwen3-VL-30B-A3B-Instruct, Avg-Bit=W32026.06 | 76.58 | |
| CLSE localBackbone=InternVL-2.5-8B, Token Reduction Ratio=66.7%, FFT Type=Local2026.06 | 75.9 | |
| CLSE globalBackbone=InternVL-2.5-8B, Token Reduction Ratio=66.7%, FFT Type=Global2026.06 | 75.9 | |
| FastVBackbone=InternVL-2.5-8B, Token Reduction Ratio=66.7%2026.06 | 75.5 | |
| MODEModel=Qwen3-VL-30B-A3B-Instruct, Avg-Bit=W22026.06 | 75.45 | |
| MC-MoEModel=Kimi-VL-A3B-Instruct, Avg-Bit=W22026.06 | 75.15 | |
| CLSE globalBackbone=InternVL-2.5-8B, Token Reduction Ratio=77.8%, FFT Type=Global2026.06 | 75 | |
| CLSE localBackbone=InternVL-2.5-8B, Token Reduction Ratio=77.8%, FFT Type=Local2026.06 | 74.9 | |
| VEQ-MEModel=Kimi-VL-A3B-Instruct, Avg-Bit=W22026.06 | 74.06 | |
| AsymVLMKeep ratio=75%, FLOPs Saved=28%2026.05 | 73.78 | |
| FastVBackbone=InternVL-2.5-8B, Token Reduction Ratio=77.8%2026.06 | 73.7 | |
| SparseVLMKeep ratio=75%, FLOPs Saved=18%2026.05 | 73.66 | |
| BaselineKeep ratio=100%, FLOPs Saved=—2026.05 | 73.54 | |
| FastVKeep ratio=75%, FLOPs Saved=0%2026.05 | 73.48 | |
| AsymVLMKeep ratio=65%, FLOPs Saved=39%2026.05 | 73.45 | |
| SparseVLMKeep ratio=65%, FLOPs Saved=35%2026.05 | 73.34 | |
| Uni-DPOModel=Qwen2-VL-2B2025.06 | 73.23 | |
| AsymVLMKeep ratio=50%, FLOPs Saved=54%2026.05 | 73.2 | |
| SparseVLMKeep ratio=50%, FLOPs Saved=47%2026.05 | 73.2 | |
| FastV+RerouteAverage Token Reduction=66.7%, Backbone=Qwen3-VL-8B2026.06 | 73 | |
| FastVKeep ratio=65%, FLOPs Saved=0%2026.05 | 72.66 | |
| CLSE localBackbone=InternVL-2.5-8B, Token Reduction Ratio=88.9%, FFT Type=Local2026.06 | 72.5 | |
| VEQ-MEModel=Qwen3-VL-30B-A3B-Instruct, Avg-Bit=W22026.06 | 72.38 | |
| FastVKeep ratio=50%, FLOPs Saved=0%2026.05 | 72.24 | |
| ILLUMEBaseline=Vicuna-7B, Params=7B2026.06 | 72.1 | |
| CLSE globalBackbone=InternVL-2.5-8B, Token Reduction Ratio=88.9%, FFT Type=Global2026.06 | 71.9 | |
| PDrop+RerouteAverage Token Reduction=77.8%, Backbone=Qwen3-VL-8B2026.06 | 71.7 | |
| PDropAverage Token Reduction=66.7%, Backbone=Qwen3-VL-8B2026.06 | 71 | |
| MC-MoEModel=Qwen3-VL-30B-A3B-Instruct, Avg-Bit=W22026.06 | 70.42 | |
| MoEQuantModel=Kimi-VL-A3B-Instruct, Avg-Bit=W22026.06 | 68.84 | |
| FastVAverage Token Reduction=77.8%, Backbone=Qwen3-VL-8B2026.06 | 68.5 | |
| Qwen2.5-VL-3BKV Size=2562026.05 | 68.2 | |
| EMU2 Chat#Params=34B, TFLOPs=5.42025.06 | 66.6 | |
| FastVBackbone=InternVL-2.5-8B, Token Reduction Ratio=88.9%2026.06 | 66.4 | |
| Cambrian-4B + Gaze attentionKV Size=144+42026.05 | 66.2 | |
| Qwen2.5-VL-3B + InfiniPot-VKV Size=1282026.05 | 66.1 | |
| Cambrian-4B + Gaze attentionKV Size=288+42026.05 | 66 | |
| LLaVA-Next-LLaMA-3-8BKV Size=7842026.05 | 64.7 | |
| EMU3#Params=8B, TFLOPs=4.12025.06 | 64.7 | |
| Emu3Baseline=Emu32026.06 | 64.7 | |
| Cambrian-4B + Gaze attentionKV Size=72+42026.05 | 64.6 | |
| Cambrian-4BKV Size=5762026.05 | 64.2 | |
| FastV+RerouteAverage Token Reduction=77.8%, Backbone=Qwen3-VL-8B2026.06 | 64.2 | |
| LLaVA-Next-Qwen2-7BKV Size=7842026.05 | 63.6 | |
| Cambrian-4B + TokenCompre. + Gaze attentionKV Size=54+42026.05 | 62.7 | |
| Speed-QModel=Qwen3-VL-30B-A3B-Instruct, Avg-Bit=W22026.06 | 62.63 | |
| Cambrian-8B-CLIP-LKV Size=5762026.05 | 62.4 | |
| Cambrian-4B + TokenCompre. + Gaze attentionKV Size=36+42026.05 | 62.3 | |
| TokenFlowBaseline=TokenFlow2026.06 | 62.3 | |
| Speed-QModel=Kimi-VL-A3B-Instruct, Avg-Bit=W22026.06 | 62.02 | |
| MoEQuantModel=Qwen3-VL-30B-A3B-Instruct, Avg-Bit=W22026.06 | 61.32 | |
| VanillaBackbone=LLaVA-1.5-13B, Retained Tokens=5762026.06 | 61.2 | |
| VILA-UBaseline=VILA-U2026.06 | 60.8 | |
| SimPOModel=Qwen2-VL-2B2025.06 | 60.63 | |
| PDrop+RerouteAverage Token Reduction=88.9%, Backbone=Qwen3-VL-8B2026.06 | 60.6 | |
| MetaMorph#Params=8B, TFLOPs=1.12025.06 | 60.5 | |
| MetaMorphBaseline=MetaMorph2026.06 | 60.5 | |
| Qwen2.5-VL-3B + InfiniPot-VKV Size=502026.05 | 60.2 | |
| Cambrian-4B + TokenCompre.KV Size=1442026.05 | 60.1 | |
| MBQModel=Kimi-VL-A3B-Instruct, Avg-Bit=W22026.06 | 59.98 | |
| MBQModel=Qwen3-VL-30B-A3B-Instruct, Avg-Bit=W22026.06 | 59.84 | |
| PDropBackbone=LLaVA-1.5-13B, Retained Tokens=1922026.06 | 59.5 | |
| VisPrunerBackbone=LLaVA-1.5-13B, Retained Tokens=1922026.06 | 59.4 | |
| PriorTRBackbone=LLaVA-1.5-13B, Retained Tokens=1922026.06 | 59.4 | |
| VisPrunerBackbone=LLaVA-1.5-13B, Retained Tokens=1282026.06 | 59.2 | |
| PriorTRBackbone=LLaVA-1.5-13B, Retained Tokens=1282026.06 | 58.4 | |
| PDropBackbone=LLaVA-1.5-13B, Retained Tokens=1282026.06 | 58.3 | |
| VanillaBackbone=LLaVA-1.5-7B, Retained Tokens=576 (Upper Bound)2026.06 | 58.2 | |
| LLaVA-OV-7B + HERMESKV Size=1002026.05 | 58 | |
| Zoo-PruneBackbone=LLaVA-1.5-7B, Retained Tokens=1282026.06 | 57.9 | |
| VisPrunerBackbone=LLaVA-1.5-13B, Retained Tokens=642026.06 | 57.7 | |
| STSBackbone=LLaVA-1.5-7B, Retained Tokens=1282026.06 | 57.6 |