Visual Question Answering on MME (Total Score)
2,448MME Total ScoreQwen2.5-VL
Evaluation Results
| Method | Links | |
|---|---|---|
| Qwen2.5-VLModel Size=72B2026.04 | 2,448 | |
| VanillaSource=Qwen’25, Backbone=Qwen3.5-9B-Hybrid, Average Gated Attention Token Reduction=0%2026.06 | 2,378 | |
| VanillaBackbone=QWEN-2.5-VL 7B, Average Tokens=100%2026.02 | 2,347 | |
| Qwen2.5-VLModel Size=7B2026.04 | 2,347 | |
| NuwaBackbone=QWEN-2.5-VL 7B, Average Tokens=75%2026.02 | 2,340 | |
| NuwaBackbone=QWEN-2.5-VL 7B, Average Tokens=50%2026.02 | 2,330 | |
| ForeSightModel Size=7B2026.04 | 2,326 | |
| VanillaSource=Qwen’25, Token Reduction=0%, Backbone=Qwen2.5-VL-7B2026.06 | 2,318 | |
| NuwaBackbone=QWEN-2.5-VL 7B, Average Tokens=25%2026.02 | 2,313 | |
| Qwen2.5-VLModel Size=7B, Reproduced=true2026.04 | 2,288.2 | |
| PDrop+RerouteSource=Ours, Token Reduction=66.7%, Backbone=Qwen2.5-VL-7B2026.06 | 2,243 | |
| PDropSource=CVPR’25, Token Reduction=66.7%, Backbone=Qwen2.5-VL-7B2026.06 | 2,213 | |
| FastV+RerouteSource=Ours, Token Reduction=66.7%, Backbone=Qwen2.5-VL-7B2026.06 | 2,184 | |
| FastVSource=ECCV’24, Token Reduction=66.7%, Backbone=Qwen2.5-VL-7B2026.06 | 2,178 | |
| PDrop+RerouteSource=Ours, Token Reduction=77.8%, Backbone=Qwen2.5-VL-7B2026.06 | 2,078 | |
| PDropSource=CVPR’25, Token Reduction=77.8%, Backbone=Qwen2.5-VL-7B2026.06 | 2,056 | |
| PDrop+RerouteSource=Ours, Backbone=Qwen3.5-9B-Hybrid, Average Gated Attention Token Reduction=66.7%2026.06 | 2,042 | |
| FastV+RerouteSource=Ours, Token Reduction=77.8%, Backbone=Qwen2.5-VL-7B2026.06 | 2,002 | |
| FastVSource=ECCV’24, Token Reduction=77.8%, Backbone=Qwen2.5-VL-7B2026.06 | 1,994 | |
| PDrop+RerouteSource=Ours, Backbone=Qwen3.5-9B-Hybrid, Average Gated Attention Token Reduction=77.8%2026.06 | 1,932 | |
| FastV+RerouteSource=Ours, Backbone=Qwen3.5-9B-Hybrid, Average Gated Attention Token Reduction=66.7%2026.06 | 1,929 | |
| PDropSource=CVPR’25, Backbone=Qwen3.5-9B-Hybrid, Average Gated Attention Token Reduction=66.7%2026.06 | 1,924 | |
| PDrop+RerouteSource=Ours, Token Reduction=88.9%, Backbone=Qwen2.5-VL-7B2026.06 | 1,874 | |
| FastVSource=ECCV’24, Backbone=Qwen3.5-9B-Hybrid, Average Gated Attention Token Reduction=66.7%2026.06 | 1,823 | |
| FastV+RerouteSource=Ours, Backbone=Qwen3.5-9B-Hybrid, Average Gated Attention Token Reduction=77.8%2026.06 | 1,798 | |
| PDropSource=CVPR’25, Token Reduction=88.9%, Backbone=Qwen2.5-VL-7B2026.06 | 1,755 | |
| PDrop+RerouteSource=Ours, Backbone=Qwen3.5-9B-Hybrid, Average Gated Attention Token Reduction=88.9%2026.06 | 1,711 | |
| PDropSource=CVPR’25, Backbone=Qwen3.5-9B-Hybrid, Average Gated Attention Token Reduction=77.8%2026.06 | 1,688 | |
| FastVSource=ECCV’24, Backbone=Qwen3.5-9B-Hybrid, Average Gated Attention Token Reduction=77.8%2026.06 | 1,547 | |
| FastV+RerouteSource=Ours, Backbone=Qwen3.5-9B-Hybrid, Average Gated Attention Token Reduction=88.9%2026.06 | 1,539 | |
| PDropSource=CVPR’25, Backbone=Qwen3.5-9B-Hybrid, Average Gated Attention Token Reduction=88.9%2026.06 | 1,431 | |
| LLaVA-1.5-13BModel=LLaVA-1.5-13B, Cropped Images=false2025.05 | 1,390 | |
| LLaVA-1.5-13B + QLIPModel=LLaVA-1.5-13B, Cropped Images=true2025.05 | 1,388 | |
| FastVSource=ECCV’24, Backbone=Qwen3.5-9B-Hybrid, Average Gated Attention Token Reduction=88.9%2026.06 | 1,263 | |
| FastV+RerouteSource=Ours, Token Reduction=88.9%, Backbone=Qwen2.5-VL-7B2026.06 | 1,247 | |
| LLaVA-1.5-7b + QLIPModel=LLaVA-1.5-7b, Cropped Images=false2025.05 | 1,241 | |
| LLaVA-1.5-7bModel=LLaVA-1.5-7b, Cropped Images=false2025.05 | 1,207 | |
| FastVSource=ECCV’24, Token Reduction=88.9%, Backbone=Qwen2.5-VL-7B2026.06 | 1,127 |