List your company
ResearchFormerly WizWand

Video Question Answering on EgoSchema subset

81AccuracyVideoHV-Agent
49.17657.43865.773.962Nov 22, 2023Apr 20, 2024Sep 17, 2024Feb 14, 2025Jul 14, 2025Dec 11, 2025May 11, 2026
Updated 5mo ago

Evaluation results

MethodLinks
VideoHV-Agentmode=Zero-Shot2026.0381
VideoAgent2mode=Zero-Shot2026.0380.6
VideoMultiAgentsmode=Zero-shot2025.0475.4
VideoMultiAgentsmode=Zero-Shot2026.0375.4
ProVCALLM/MLLM=GPT-4o2026.0474.2
InternVL-3 + A.I.R.VLM Size=8B, #Frames=≤ 322025.1072.2
Lifelong Memorymode=Zero-shot2025.0472
LifelongMemorymode=Zero-Shot2026.0372
InternVL-3VLM Size=8B, #Frames=322025.1071.6
InternVL-3 + MDP3VLM Size=8B, #Frames=322025.1070
AKEYS(M)LLM=GPT-4o2025.0368.6
Tarsiermode=Zero-shot2025.0468.6
Tarsiermode=Zero-Shot2026.0368.6
LVNet(M)LLM=GPT-4o2025.0368.2
LVNetmode=Zero-shot2025.0468.2
LVNet(M)LLM=GPT-4o2024.0568.2
LVNetmode=Zero-Shot2026.0368.2
LVNetLLM/MLLM=GPT-4o2026.0468.2
AKEYS(M)LLM=GPT-42025.0368
Qwen2-VL-InstructLLM=Qwen2-7B, Vision Encoder=DFN-CLIP-H2024.1266.7
T*VLM Size=7B, #Frames=82025.1066.6
DrVideoVLM Size=GPT42025.1066.4
DrVideoLLM/MLLM=GPT-42026.0466.4
LangRepo(M)LLM=Mixtral-8x7B2025.0366.2
VideoTree(M)LLM=GPT-42025.0366.2
VideoTreemode=Zero-shot2025.0466.2
LangRepo(M)LLM=Mixtral-8x7B2024.0566.2
VIDEOTREE(M)LLM=GPT-42024.0566.2
VideoTreemode=Zero-Shot2026.0366.2
VideoTreeVLM Size=GPT42025.1066.2
LangRepoLLM/MLLM=Mixtral-8×7B2026.0466.2
VideoTreeLLM/MLLM=GPT-42026.0466.2
Hu et al. (2025)VLM Size=8.5B, #Frames=322025.1065.9
LifelongMemory(M)LLM=GPT-42025.0364.1
LifelongMemory(M)LLM=GPT-42024.0564.1
LLaVA-OneVision 32 frames + BOLTLLM Size=7B, Training Free=true, Frames=322025.0364
LLaVA-OneVision + BOLTVLM Size=7B, #Frames=322025.1064
VideoAgentcitation=[9], (M)LLM=GPT-42024.0563.8
VideoChat2LLM Size=7B, Training Free=false2025.0363.6
LLaVA-OneVision + A.I.R.VLM Size=7B, #Frames=≤ 322025.1063.2
MI-PrunerBackbone=Qwen2.5VL-7B, Frames / Pruning Configuration=F160 (5×) / 20%2026.0563
VideoAgent [6](M)LLM=GPT-42025.0362.8
VideoAgentLLM/MLLM=GPT-4, Reference=[28]2026.0462.8
QwenVL-2.5 + A.I.R.VLM Size=7B, #Frames=≤ 322025.1062.4
LLaVA-OneVision 8 frames + BOLTLLM Size=7B, Training Free=true, Frames=82025.0362.2
LLaVA-OneVision 32 framesLLM Size=7B, Training Free=false, Frames=322025.0362.2
LLaVA-OneVision 8 framesLLM Size=7B, Training Free=false, Frames=82025.0362
LLaVA-OneVision 16 frames + BOLTLLM Size=7B, Training Free=true, Frames=162025.0361.8
LLaVA-OneVisionVLM Size=7B, #Frames=322025.1061.8
QwenVL-2.5 + MDP3VLM Size=7B, #Frames=322025.1061.6
LLaVA-OneVision 16 framesLLM Size=7B, Training Free=false, Frames=162025.0361.4
LLoVi(M)LLM=GPT-42025.0361.2
LLoVi(M)LLM=GPT-42024.0561.2
LLoViLLM/MLLM=GPT-42026.0461.2
EchoPruneBackbone=Qwen2.5VL-7B, Frames / Pruning Configuration=F160 (5×) / 20%2026.0561.2
RandomBackbone=Qwen2.5VL-7B, Frames / Pruning Configuration=F160 (5×) / 20%2026.0561
FlashVIDBackbone=Qwen2.5VL-7B, Frames / Pruning Configuration=F160 (5×) / 20%2026.0561
LLaVA-OneVision + MDP3VLM Size=7B, #Frames=322025.1060.8
Qwen2.5VL-7BBackbone=Qwen2.5VL-7B, Frames / Pruning Configuration=F32 (1×) / 100%2026.0560.6
MVU(M)LLM=Mistral-13B2025.0360.3
MVU(M)LLM=Mistral-13B2024.0560.3
MVUVLM Size=13B, #Frames=162025.1060.3
MVULLM/MLLM=Mistral-13B2026.0460.3
VideoAgent [38](M)LLM=GPT-42025.0360.2
VideoAgentcitation=[67], (M)LLM=GPT-42024.0560.2
VideoAgentmode=Zero-Shot2026.0360.2
VideoAgentVLM Size=GPT4, #Frames=1FPS2025.1060.2
VideoAgentLLM/MLLM=GPT-4, Reference=[4]2026.0460.2
LLaVAOneVision-7BTraining Mode=SFT, Video Data=true, Model Size=7B2024.0960.1
LLAVA-OVLLM=Qwen2-7B, Vision Encoder=SigLIP-SO400M2024.1260.1
IG-VLM(M)LLM=GPT-4V2025.0359.8
IG-VLM(M)LLM=GPT-4V2024.0559.8
IG-VLMLLM/MLLM=GPT-4V2026.0459.8
QwenVL-2.5VLM Size=7B, #Frames=322025.1059.4
HCQAmode=Zero-shot2025.0458.8
HCQAmode=Zero-Shot2026.0358.8
LLaVA-NeXT-INST-ITLLM=Vicuna-7B, Vision Encoder=CLIP-ViT-Large2024.1257.8
TS-LLAVAmode=Zero-Shot2026.0357.8
LLoVi(M)LLM=GPT-3.52024.0557.6
LLaVA-VideoLLM=Qwen2-7B, Vision Encoder=SigLIP-SO400M2024.1257.3
Vamos (GPT-4o)Input Type=caption, LLM=GPT-4o, Zero-shot=true2023.1157.2
MM1.5-Video-7BTraining Mode=SFT, Video Data=true, Model Size=7B2024.0957.2
ProViQ(M)LLM=GPT-3.52025.0357.1
ProViQ(M)LLM=GPT-3.52024.0557.1
FlashVIDBackbone=Qwen2.5VL-3B, Frames / Pruning Configuration=F160 (5×) / 20%2026.0556.2
MI-PrunerBackbone=Qwen2.5VL-3B, Frames / Pruning Configuration=F160 (5×) / 20%2026.0555.8
EchoPruneBackbone=Qwen2.5VL-3B, Frames / Pruning Configuration=F160 (5×) / 20%2026.0555.4
RandomBackbone=Qwen2.5VL-3B, Frames / Pruning Configuration=F160 (5×) / 20%2026.0555.2
Qwen2.5VL-3BBackbone=Qwen2.5VL-3B, Frames / Pruning Configuration=F32 (1×) / 100%2026.0553.8
VILA-1.5 + Frame-VoyagerVLM Size=8B, #Frames=82025.1053.6
VILA-1.5 + A.I.R.VLM Size=8B, #Frames=82025.1053.6
VILA-1.5VLM Size=8B, #Frames=82025.1052.8
MM1.5-Video-3BTraining Mode=SFT, Video Data=true, Model Size=3B2024.0952.4
Video-LLaMA2-7BTraining Mode=SFT, Video Data=true, Model Size=7B2024.0951.7
Vamos (GPT-4)Input Type=caption, LLM=GPT-4, Zero-shot=true2023.1151.2
Vamos(M)LLM=GPT-42025.0351.2
Vamos(M)LLM=GPT-42024.0551.2
MM1.5-Video-1BTraining Mode=SFT, Video Data=true, Model Size=1B2024.0951
VILA-1.5 + MDP3VLM Size=8B, #Frames=82025.1051
LLaVA-NeXT-INST-ITLLM=Qwen2-7B, Vision Encoder=SigLIP-SO4002024.1250.4