List your company

Video Question Answering on VideoMME (wo/w-subs)

75VQA Score (wo subs)Gemini-1.5-Pro
37.35247.12656.966.674Nov 22, 2024Feb 13, 2025May 8, 2025Jul 31, 2025Oct 23, 2025Jan 15, 2026Apr 9, 2026

Evaluation results

MethodLinks
Gemini-1.5-ProFrames=568, Inference Time=198.8s2025.067581.3—
VideoChat-A1Backbone=InternVideo2.5-8B, Frames=41.3, Inference Time=28.4s2025.0672.976.8—
VideoChat-A1Backbone=InternVL2.5-8B, Frames=35.9, Inference Time=14.7s2025.0672.377—
GPT-4oFrames=384, Inference Time=134.4s2025.0671.977.2—
VideoChat-A1Backbone=Qwen2.5-VL-7B, Frames=42.0, Inference Time=18.6s2025.0671.876.5—
Qwen2.5-VL-72BFrames=768, Inference Time=122.4s2025.0671.277.8—
Bridge-STGModel Scale=7B-based2026.0467.974.8—
DyCokeBackbone=LLaVA-OV-72B, K=0.5, L=3, P=0.72024.1166.369.7—
OmniZipBackbone=Qwen2.5-Omni-7B, Retained Ratio=45%, FLOPs Ratio=39%2025.1166.3—63
Full TokensBackbone=LLaVA-OV-72B2024.1166.269.5—
OmniZipBackbone=Qwen2.5-Omni-7B, Retained Ratio=35%, FLOPs Ratio=29%2025.1166.1—61
Full TokensBackbone=Qwen2.5-Omni-7B, Retained Ratio=100%, FLOPs Ratio=100%2025.1166—64.5
DyCoke (V&A)Backbone=Qwen2.5-Omni-7B, Retained Ratio=50%, FLOPs Ratio=44%2025.1165.5—62
RandomBackbone=Qwen2.5-Omni-7B, Retained Ratio=55%, FLOPs Ratio=48%2025.1165.4—61
DyCoke (V&A)Backbone=Qwen2.5-Omni-7B, Retained Ratio=35%, FLOPs Ratio=29%2025.1165.2—57.4
InternVideo-2.5-8BFrames=512, Inference Time=33.8s2025.0665.1——
RandomBackbone=Qwen2.5-Omni-7B, Retained Ratio=40%, FLOPs Ratio=34%2025.1165—56.9
DyCokeBackbone=LLaVA-OV-72B, K=0.7, L=3, P=0.72024.1164.968.7—
InternVL-2.5-8BFrames=64, Inference Time=12.4s2025.0664.266.9—
Qwen2.5-VL-7BFrames=512, Inference Time=24.3s2025.0663.369.7—
OmniZipBackbone=Qwen2.5-Omni-3B, Retained Ratio=45%, FLOPs Ratio=36%2025.1162.8—61.3
OmniZipBackbone=Qwen2.5-Omni-3B, Retained Ratio=35%, FLOPs Ratio=26%2025.1162.7—60.1
Full TokensBackbone=Qwen2.5-Omni-3B, Retained Ratio=100%, FLOPs Ratio=100%2025.1162.6—62.2
FastVBackbone=LLaVA-OV-72B2024.1162.166.7—
PruMergeBackbone=LLaVA-OV-72B2024.1161.963.8—
DyCoke (V&A)Backbone=Qwen2.5-Omni-3B, Retained Ratio=50%, FLOPs Ratio=40%2025.1161.6—60.7
RandomBackbone=Qwen2.5-Omni-3B, Retained Ratio=55%, FLOPs Ratio=45%2025.1161.1—58.7
DyCoke (V&A)Backbone=Qwen2.5-Omni-3B, Retained Ratio=35%, FLOPs Ratio=26%2025.1161—57.4
RandomBackbone=Qwen2.5-Omni-3B, Retained Ratio=40%, FLOPs Ratio=31%2025.1160.6—57.5
SpaceVLLMModel Scale=7B-based2026.046065.6—
DyCokeBackbone=LLaVA-OV-7B, K=0.5, L=3, P=0.72024.1159.561.4—
DyCokeBackbone=LLaVA-OV-7B, K=0.3, L=3, P=0.72024.1158.861—
Full TokensBackbone=LLaVA-OV-7B2024.1158.561.3—
DyCokeBackbone=LLaVA-OV-7B, K=0.7, L=3, P=0.72024.1158.360.7—
LLaVA-OVModel Scale=7B-based2026.0458.261.5—
FastVBackbone=LLaVA-OV-7B2024.1157.360.5—
Videollama2.1Model Scale=7B-based2026.0454.956.4—
PruMergeBackbone=LLaVA-OV-7B2024.1152.957—
DyCokeBackbone=LLaVA-OV-0.5B, K=0.3, L=3, P=0.72024.1145.443.8—
DyCokeBackbone=LLaVA-OV-0.5B, K=0.7, L=3, P=0.72024.1145.243.3—
DyCokeBackbone=LLaVA-OV-0.5B, K=0.5, L=3, P=0.72024.1145.143.4—
Full TokensBackbone=LLaVA-OV-0.5B2024.1144.143.5—
FastVBackbone=LLaVA-OV-0.5B2024.114241.4—
Video-LLaVAModel Scale=7B-based2026.0439.941.6—
PruMergeBackbone=LLaVA-OV-0.5B2024.1138.839.3—
FastVBackbone=Qwen2.5-Omni-7B, Retained Ratio=50%, FLOPs Ratio=54%2025.11——58.4
FastVBackbone=Qwen2.5-Omni-7B, Retained Ratio=35%, FLOPs Ratio=42%2025.11——57.8
FastVBackbone=Qwen2.5-Omni-3B, Retained Ratio=50%, FLOPs Ratio=49%2025.11——55.9
FastVBackbone=Qwen2.5-Omni-3B, Retained Ratio=35%, FLOPs Ratio=37%2025.11——57.7