Multimodal Perception and Cognition on MME
2,515.82Overall ScoreVanilla
Evaluation results
| Method | Links | |||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| VanillaBackbone=Qwen2.5-VL-7B, TFLOPs=100%2026.06 | 2,515.82 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Qwen3-VL-32BToken Ratio=100%2026.05 | 2,501.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| F3AToken Ratio=60%2026.05 | 2,494.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Qwen3-VL-8BRetained Tokens=1024 Tokens (100%)2026.07 | 2,440.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CDPrunerToken Ratio=60%2026.05 | 2,436.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SophiaVL-R1-7BParameters=7B2025.05 | 2,403.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VisionZipToken Ratio=60%2026.05 | 2,402.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VanillaBackbone=Qwen3-VL-8B, TFLOPs=100%2026.06 | 2,390.38 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VisionZipToken Ratio=40%2026.05 | 2,381.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FastVToken Ratio=60%2026.05 | 2,380.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CDPrunerToken Ratio=40%2026.05 | 2,376.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DivPruneToken Ratio=60%2026.05 | 2,370.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| F3AToken Ratio=40%2026.05 | 2,355.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Qwen2.5-VL-7B-InstructParameters=7B, Training Strategy=SFT+GRPO2025.05 | 2,344.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Qwen2.5-VL-7BBase Model=Qwen2.5-VL-7B, Attention Skipping Method=None2026.06 | 2,344.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Op-SkipBackbone=Qwen3-VL-8B, TFLOPs=66%, N=202026.06 | 2,336.01 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| APETBackbone=Qwen3-VL-8B, TFLOPs=66%2026.06 | 2,332.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GPT-4oModel Modality Type=Proprietary2024.10 | 2,328.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Qwen2 VLModel Scale=7B, Model Modality Type=Vision-language2024.10 | 2,326.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VTWBase Model=Qwen2.5-VL-7B, Attention Skipping Method=VTW, Hyperparameter (K/N)=K=72026.06 | 2,319.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PIO-FVLMToken Budget=Retain 33.30% Tokens, Venue=Ours, Backbone=Qwen-2.5-VL-7B2026.02 | 2,317 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DARTToken Budget=Retain 33.30% Tokens, Venue=EMNLP'25, Backbone=Qwen-2.5-VL-7B2026.02 | 2,316 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Qwen2.5-VL-7B-InstructParameters=7B, Training Strategy=Direct2025.05 | 2,306 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VanillaToken Budget=1296 Tokens (Upper Bound), Backbone=Qwen-2.5-VL-7B2026.02 | 2,299 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Qwen2.5-VL-7B-InstructParameters=7B, Training Strategy=GRPO2025.05 | 2,298.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| V2DropBackbone=Qwen3-VL-8B, TFLOPs=66%2026.06 | 2,290.27 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FP16Model=Qwen2VL-7B, Spiking=false, Time Step=N/A2026.04 | 2,285 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PIO-FVLMToken Budget=Retain 22.2% Tokens, Venue=Ours, Backbone=Qwen-2.5-VL-7B2026.02 | 2,284 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FastVToken Ratio=40%2026.05 | 2,273 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| F3AToken Ratio=20%2026.05 | 2,269.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| APETBackbone=Qwen2.5-VL-7B, TFLOPs=55%2026.06 | 2,263.83 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Op-SkipBackbone=Qwen2.5-VL-7B, TFLOPs=55%, N=202026.06 | 2,263.38 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ShortVBackbone=Qwen3-VL-8B, TFLOPs=58%, N=202026.06 | 2,263.21 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EADPRetained Tokens=512 Tokens (↓ 50.0%)2026.07 | 2,261.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LLaVA-OneVision-72BParameters=72B2025.05 | 2,261 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| V2DropBackbone=Qwen2.5-VL-7B, TFLOPs=55%2026.06 | 2,253.59 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SpikeMLLM(QuaRot+MSTS+TC-LIF)Model=Qwen2VL-7B, Spiking=true, Time Step=3/42026.04 | 2,236 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DivPruneToken Ratio=40%2026.05 | 2,234.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| IXC-2.5-7BModel Size=7B2024.07 | 2,229 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| V-SkipBase Model=Qwen2.5-VL-7B, Attention Skipping Method=V-Skip, Hyperparameter (K/N)=N=72026.06 | 2,228.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DARTToken Budget=Retain 22.2% Tokens, Venue=EMNLP'25, Backbone=Qwen-2.5-VL-7B2026.02 | 2,227 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VSkipBackbone=Qwen3-VL-8B, TFLOPs=76%, N=202026.06 | 2,226.98 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| R1-Onevision-7BParameters=7B2025.05 | 2,223.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FP16Model=MiniCPM-V-2.6-8B, Spiking=false, Time Step=N/A2026.04 | 2,223 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DivPruneRetained Tokens=512 Tokens (↓ 50.0%)2026.07 | 2,217.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FP16Model=InternVL2-8B, Spiking=false, Time Step=N/A2026.04 | 2,211 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EADPRetained Tokens=256 Tokens (↓ 75.0%)2026.07 | 2,202.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CDPrunerRetained Tokens=256 Tokens (↓ 75.0%)2026.07 | 2,200.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EADPRetained Tokens=128 Tokens (↓ 87.5%)2026.07 | 2,200.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CDPrunerRetained Tokens=512 Tokens (↓ 50.0%)2026.07 | 2,199.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CDPrunerToken Ratio=20%2026.05 | 2,198.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SpikeMLLM(QuaRot+MSTS+TC-LIF)Model=InternVL2-8B, Spiking=true, Time Step=3/42026.04 | 2,198 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SpikeMLLM(QuaRot+MSTS+TC-LIF)Model=MiniCPM-V-2.6-8B, Spiking=true, Time Step=3/42026.04 | 2,197 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SpikeMLLM(QuaRot)Model=Qwen2VL-7B, Spiking=true, Time Step=152026.04 | 2,196 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| InternVLModel Size=1.5-26B, Access Type=Open-source2024.07 | 2,189.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VITAModel Scale=8x7B, Model Modality Type=Omni-modal2024.10 | 2,189.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VisionZipToken Ratio=20%2026.05 | 2,189.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Baichuan-omniModel Scale=7B, Model Modality Type=Omni-modal2024.10 | 2,186.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DivPruneRetained Tokens=256 Tokens (↓ 75.0%)2026.07 | 2,185.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SpikeMLLM(QuaRot)Model=InternVL2-8B, Spiking=true, Time Step=152026.04 | 2,157 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CDPrunerRetained Tokens=128 Tokens (↓ 87.5%)2026.07 | 2,153.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PIO-FVLMToken Budget=Retain 11.1% Tokens, Venue=Ours, Backbone=Qwen-2.5-VL-7B2026.02 | 2,150 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Gemini Pro2024.08 | 2,148.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DivPruneRetained Tokens=128 Tokens (↓ 87.5%)2026.07 | 2,136.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ShortVBase Model=Qwen2.5-VL-7B, Attention Skipping Method=ShortV, Hyperparameter (K/N)=N=72026.06 | 2,092.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SpikeMLLM(GPTQ)Model=MiniCPM-V-2.6-8B, Spiking=true, Time Step=2552026.04 | 2,076 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SpikeMLLM(QuaRot)Model=MiniCPM-V-2.6-8B, Spiking=true, Time Step=152026.04 | 2,069 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DivPruneToken Ratio=20%2026.05 | 2,059.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SpikeMLLM(QuaRot+MSTS+TC-LIF)Model=InternVL2-8B, Spiking=true, Time Step=2/32026.04 | 2,058 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Yi-VL-34BSize=34B2024.08 | 2,050.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Cambrian-34BSize=34B2024.08 | 2,049.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Phantom-3.8BModel Size=3.8B2024.09 | 2,046 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VSkipBackbone=Qwen2.5-VL-7B, TFLOPs=67%, N=162026.06 | 2,039.69 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TroL-1.8BModel Size=1.8B2024.09 | 2,038 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SpikeMLLM(GPTQ)Model=Qwen2VL-7B, Spiking=true, Time Step=2552026.04 | 2,032 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MiniCPM-Llama3-V 2.5Size=8.5B2024.08 | 2,024.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MiniCPM-Llama3-V 2.5Model Scale=8B, Model Modality Type=Vision-language2024.10 | 2,024.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FastVToken Ratio=20%2026.05 | 2,023.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LLaVA-NeXT-Yi-34BSize=34B2024.08 | 2,006.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GPT-4o-miniModel Modality Type=Proprietary2024.10 | 2,003.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SpikeMLLM(QuaRot+MSTS+TC-LIF)Model=MiniCPM-V-2.6-8B, Spiking=true, Time Step=2/32026.04 | 2,002 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LLaVA-OneVision-7BParameters=7B2025.05 | 1,998 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SpikeMLLM(RTN)Model=MiniCPM-V-2.6-8B, Spiking=true, Time Step=2552026.04 | 1,989 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SpikeMLLM(QuaRot+MSTS+TC-LIF)Model=Qwen2VL-7B, Spiking=true, Time Step=2/32026.04 | 1,984 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TroL-3.8BModel Size=3.8B2024.09 | 1,980 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LLaVA-NeXT-Llama3-8BSize=8.4B2024.08 | 1,971.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DARTToken Budget=Retain 11.1% Tokens, Venue=EMNLP'25, Backbone=Qwen-2.5-VL-7B2026.02 | 1,971 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Gemini-ProAccess Type=Closed-source API2024.07 | 1,933.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GPT-4VAccess Type=Closed-source API2024.07 | 1,926.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GPT-4V2025.05 | 1,926 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Bunny-Llama-3-8BSize=8.4B2024.08 | 1,920.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Yi-VL-6BSize=6.7B2024.08 | 1,915.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SpikeMLLM(RTN)Model=Qwen2VL-7B, Spiking=true, Time Step=2552026.04 | 1,904 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ShortVBase Model=LLaVA-NeXT-13B, Attention Skipping Method=ShortV, Hyperparameter (K/N)=N=242026.06 | 1,899.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LLaVA-NeXT-13BBase Model=LLaVA-NeXT-13B, Attention Skipping Method=None2026.06 | 1,892 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| V-SkipBase Model=LLaVA-NeXT-13B, Attention Skipping Method=V-Skip, Hyperparameter (K/N)=N=242026.06 | 1,885.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Phantom-1.8BModel Size=1.8B2024.09 | 1,885 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VTWBase Model=LLaVA-NeXT-13B, Attention Skipping Method=VTW, Hyperparameter (K/N)=K=202026.06 | 1,878.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| InternVL2Parameters=2B2026.03 | 1,876.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SpikeMLLM(GPTQ)Model=InternVL2-8B, Spiking=true, Time Step=2552026.04 | 1,870 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — |