List your company
ResearchFormerly WizWand

Video Question Answering on NExT-QA Multi-choice

84.5AccuracyVideoLLaMA3-7B
53.61261.63169.6577.669Sep 14, 2023Feb 12, 2024Jul 13, 2024Dec 12, 2024May 12, 2025Oct 11, 2025Mar 12, 2026
Updated 7mo ago

Evaluation results

MethodLinks
VideoLLaMA3-7BMax #F=180, Max Res.=3842026.0384.5
LLaVA-VideoLLM=Qwen2-7B, Vision Encoder=SigLIP-SO400M2024.1283.2
LLaVA-Video-7Bzero-shot=true2025.0383.2
LLaVAction-7Bzero-shot=true2025.0382.8
NVILA-8B-Video + AutoGazeMax #F=1024, Max Res.=35842026.0382.8
NVILA-8B-VideoMax #F=256, Max Res.=4482026.0382.2
LongVILA-R1-7BMax #F=8192, Max Res.=4482026.0381.5
LongVILALLM Size=7B2024.0880.7
LongVILA-7BMax #F=2048, Max Res.=3842026.0380.7
VideoChat2-7BTotal Params=7.3B, Input Res.=224×224, Num Frames=16, Throughput=11.42025.0379.5
LLAVA-OVLLM=Qwen2-7B, Vision Encoder=SigLIP-SO400M2024.1279.4
LLAVA-OVLLM Size=7B2024.0879.4
LLaVA-OV-7Bzero-shot=true2025.0379.4
LLaVA-OV-8BMax #F=32, Max Res.=3842026.0379.4
Tarsier-34BMode=Zero-shot2024.0679.2
InternVL2.5-2BTotal Params=2.2B, Input Res.=448×448, Num Frames=32, Throughput=21.82025.0375.6
Mobile-VideoGPT-1.5BTotal Params=1.6B, Input Res.=224×224, Num Frames=16, Throughput=41.02025.0373.7
TS-LLaVALLM Size=34B, Vision Encoder=CLIP-L, Training Strategy=Training-free2024.1173.6
TS-LLaVALLM Size=34B, Vision Encoder=CLIP-L, Training Strategy=Training-free2024.1173.6
TS-LLaVALLM Size=34B, Vision Encoder=CLIP-L, Training Strategy=Training-free2024.1173.6
VideoTreeLLM Size=GPT-4, Vision Encoder=Unknown, Training Protocol=Training-free2024.0773.5
VideoTreeMode=Zero-shot2024.0673.5
VideoTreeLLM Size=GPT-4, Vision Encoder=Unknown, Training Strategy=Training-free2024.1173.5
VideoTreeLLM Size=GPT-4, Vision Encoder=Unknown, Training Strategy=Training-free2024.1173.5
VideoTreeLLM Size=GPT-4, Vision Encoder=Unknown, Training Strategy=Training-free2024.1173.5
LLaVA-NeXT-INST-ITLLM=Qwen2-7B, Vision Encoder=SigLIP-SO4002024.1273
DYTOLLM Size=34B, Vis Encoder=CLIP-L, Training Regime=Training-Free2026.0272.9
KTV-34B-denseLLM Size=34B, Vis Encoder=CLIP-L, Training Regime=Training-Free, Variant=dense2026.0272.7
KTV-34B-normalLLM Size=34B, Vis Encoder=CLIP-L, Training Regime=Training-Free, Variant=normal2026.0272.3
SF-LLaVA-34BLLM Size=34B, Vision Encoder=CLIP-L, Training Protocol=Training-free2024.0772
SF-LLAVALLM Size=34B, Vision Encoder=CLIP-L, Training Strategy=Training-free2024.1172
SF-LLAVALLM Size=34B, Vision Encoder=CLIP-L, Training Strategy=Training-free2024.1172
SF-LLAVALLM Size=34B, Vision Encoder=CLIP-L, Training Strategy=Training-free2024.1172
Tarsier-7BMode=Zero-shot2024.0671.6
VideoAgentLLM Size=GPT-4, Vision Encoder=Unknown, Training Protocol=Training-free2024.0771.3
VideoAgentMode=Zero-shot2024.0671.3
VideoAgentLLM Size=GPT-4, Vision Encoder=Unknown, Training Strategy=Training-free2024.1171.3
VideoAgentLLM Size=GPT-4, Vision Encoder=Unknown, Training Strategy=Training-free2024.1171.3
VideoAgentLLM Size=GPT-4, Vision Encoder=Unknown, Training Strategy=Training-free2024.1171.3
KTV-34B-sparseLLM Size=34B, Vis Encoder=CLIP-L, Training Regime=Training-Free, Variant=sparse2026.0271.2
IG-VLM (LLaVA-v1.6)LLM Size=34B, Vision Encoder=CLIP-L, Training Protocol=Training-free2024.0770.9
IG-VLMMode=Zero-shot2024.0670.9
IG-VLMLLM Size=34B, Vision Encoder=CLIP-L, Training Strategy=Training-free2024.1170.9
IG-VLMLLM Size=34B, Vision Encoder=CLIP-L, Training Strategy=Training-free2024.1170.9
IG-VLMLLM Size=34B, Vision Encoder=CLIP-L, Training Strategy=Training-free2024.1170.9
SF-LLaVA-7BLLM Size=34B, Vis Encoder=CLIP-L, Training Regime=Training-Free2026.0270.9
IG-VLMLLM Size=34B, Vis Encoder=CLIP-L, Training Regime=Training-Free2026.0270.7
LLaVA-NeXT-INST-ITLLM=Vicuna-7B, Vision Encoder=CLIP-ViT-Large2024.1270.2
MMICLbackbone=Instruct-FLAN-T5-XXL2023.0968.8
LongVA-7Bzero-shot=true2025.0368.3
LongVA-7BTotal Params=7.4B, Input Res.=224×224, Num Frames=8, Throughput=9.22025.0368.3
LLoViLLM Size=GPT-3.5, Vision Encoder=Unknown, Training Protocol=Training-free2024.0767.7
LLoViMode=Zero-shot2024.0667.7
LLoViLLM Size=GPT-3.5, Vision Encoder=Unknown, Training Strategy=Training-free2024.1167.7
LLoViLLM Size=GPT-3.5, Vision Encoder=Unknown, Training Strategy=Training-free2024.1167.7
LLoViLLM Size=GPT-3.5, Vision Encoder=Unknown, Training Strategy=Training-free2024.1167.7
GPT-4ozero-shot=true2025.0366.7
TS-LLaVALLM Size=7B, Vision Encoder=CLIP-L, Training Strategy=Training-free2024.1166.5
TS-LLaVALLM Size=7B, Vision Encoder=CLIP-L, Training Strategy=Training-free2024.1166.5
MMICLbackbone=FLAN-T5-XL2023.0966.17
KTV-7B-denseLLM Size=7B, Vis Encoder=CLIP-L, Training Regime=Training-Free, Variant=dense2026.0265.8
DYTOLLM Size=7B, Vis Encoder=CLIP-L, Training Regime=Training-Free2026.0265.7
Mobile-VideoGPT-0.5BTotal Params=0.6B, Input Res.=224×224, Num Frames=16, Throughput=45.92025.0365.4
MMICLbackbone=Instruct-FLAN-T5-XL2023.0965.33
KTV-7B-normalLLM Size=7B, Vis Encoder=CLIP-L, Training Regime=Training-Free, Variant=normal2026.0265.1
MMICLbackbone=FLAN-T5-XXL2023.0964.67
KTV-7B-sparseLLM Size=7B, Vis Encoder=CLIP-L, Training Regime=Training-Free, Variant=sparse2026.0264.5
InstructBLIPbackbone=FLANT5-XXL2023.0964.27
SF-LLaVA-7BLLM Size=7B, Vision Encoder=CLIP-L, Training Protocol=Training-free2024.0764.2
SF-LLaVALLM Size=7B, Vision Encoder=CLIP-L, Training Strategy=Training-free2024.1164.2
SF-LLaVALLM Size=7B, Vision Encoder=CLIP-L, Training Strategy=Training-free2024.1164.2
Gemini-1.5-Prozero-shot=true2025.0364
SF-LLaVA-7BLLM Size=7B, Vis Encoder=CLIP-L, Training Regime=Training-Free2026.0264
IG-VLM (LLaVA-v1.6)LLM Size=7B, Vision Encoder=CLIP-L, Training Protocol=Training-free2024.0763.1
M3LLM Size=7B, Vision Encoder=CLIP-L, Training Strategy=Video-trained2024.1163.1
IG-VLMLLM Size=7B, Vision Encoder=CLIP-L, Training Strategy=Training-free2024.1163.1
M3LLM Size=7B, Vision Encoder=CLIP-L, Training Strategy=Video-trained2024.1163.1
IG-VLMLLM Size=7B, Vision Encoder=CLIP-L, Training Strategy=Training-free2024.1163.1
IG-VLMLLM Size=7B, Vis Encoder=CLIP-L, Training Regime=Training-Free2026.0263.1
BLIP-2backbone=FLANT5-XXL2023.0961.97
BLIP-2backbone=FLANT5-XL2023.0961.73
Flash-VStreamLLM Size=7B2024.0861.6
Gemini-1.5-Flashzero-shot=true2025.0361.6
GPT-4Vzero-shot=true2025.0361.3
DeepStack-LLLM Size=7B, Vision Encoder=CLIP-L, Training Protocol=SFT2024.0761
DeepStack-LLLM Size=7B, Vision Encoder=CLIP-L, Training Strategy=Video-trained2024.1161
DeepStack-LLLM Size=7B, Vision Encoder=CLIP-L, Training Strategy=Video-trained2024.1161
DeepStack-LLLM=Vicuna-7B, Vision Encoder=CLIP-ViT-Large2024.1261
LangRepoMode=Zero-shot2024.0660.9
Vista-LLaMALLM Size=7B, Vision Encoder=CLIP-G, Training Protocol=SFT2024.0760.7
Vista-LLaMALLM Size=7B, Vision Encoder=CLIP-G, Training Strategy=Video-trained2024.1160.7
Vista-LLaMALLM Size=7B, Vision Encoder=CLIP-G, Training Strategy=Video-trained2024.1160.7
Video-LLaVALLM Size=7B, Vision Encoder=ViT-L, Training Strategy=Video-trained2024.1160.5
Video-LLaVALLM Size=7B, Vision Encoder=ViT-L, Training Strategy=Video-trained2024.1160.5
LLaVA-NeXTLLM=Vicuna-7B, Vision Encoder=CLIP-ViT-Large, reproduced=true2024.1258.4
Full ModelSparsity Ratio=0%, Base Model=LLaVA-OneVision2025.0458.4
LLaVA-OneVision-0.5BTotal Params=1.0B, Input Res.=384×384, Num Frames=32, Throughput=22.72025.0357.2
MovieChat+LLM Size=7B, Vision Encoder=CLIP-G, Training Protocol=SFT2024.0754.8
MovieChat+LLM Size=7B, Vision Encoder=CLIP-G, Training Strategy=Video-trained2024.1154.8
MovieChat+LLM Size=7B, Vision Encoder=CLIP-G, Training Strategy=Video-trained2024.1154.8