List your company
ResearchFormerly WizWand

Long-form Video Understanding on LongVideoBench (Accuracy)

82.3AccuracyVideoChat-M1
54.53261.74168.9576.159Sep 19, 2024Dec 28, 2024Apr 8, 2025Jul 17, 2025Oct 26, 2025Feb 3, 2026May 15, 2026
Updated 5mo ago

Evaluation results

MethodLinks
VideoChat-M1Model Category=Ours, Number of Parameters=37B2025.1182.3
GPT-4oAgent=disabled2026.0581.3
Gemini 2.5 ProModel Category=Closed-source Large-sized MLLM2025.1178.7
Seed1.5-VLThinking mode=false, Evaluation frame rate=1 FPS2025.0574.4
Seed1.5-VLThinking mode=true, Evaluation frame rate=1 FPS2025.0574
Seed 1.5VLModel Category=Closed-source Large-sized MLLM2025.1174
DeepVideoDiscoveryModel Category=Agent-based Methods2025.1171.6
NVILAModel Size=8B, Video Tokens=8K2025.1170.1
InternVL3.5-241B-A28B2026.0167.1
InternVL-3.5-241BModel Category=Open-source Large-sized MLLM2025.1167.1
GPT-4o2024.0966.7
GPT-4oEvaluation frame rate=1 FPS2025.0566.7
GPT-4oSize=-2026.0266.7
GPT-4oCategory=Proprietary Models2025.1266.7
GPT-4oModel Category=Proprietary Models2025.1266.7
GPT-4oModel Category=Closed-source Large-sized MLLM2025.1166.7
GPT4o2025.1166.7
VideoSeeker-8BAgent=enabled2026.0566.5
Eagle-2.5-8BModel Category=Open-source Medium-sized MLLM2025.1166.4
InternVL-3-78BModel Category=Open-source Large-sized MLLM2025.1165.7
InternVL-3.5-78BModel Category=Open-source Large-sized MLLM2025.1165.7
VideoChat-A1Model Category=Agent-based Methods2025.1165.4
VideoRAG-72BModel Category=Agent-based Methods2025.1165.4
LLAVA-Video-72BModel Category=Open-source Large-sized MLLM2025.1164.9
VideoChat-Flash-7BModel Category=Open-source Medium-sized MLLM2025.1164.7
Qwen3-VL-8BAgent=disabled2026.0564.6
Gemini-1.5-Pro2024.0964.4
Aria-28BModel Category=Open-source Large-sized MLLM2025.1164.2
VideoSeeker-4BAgent=enabled2026.0564.2
Gemini-1.5-ProSize=-2026.0264
Gemini-1.5-proCategory=Proprietary Models2025.1264
Gemini-1.5-proModel Category=Proprietary Models2025.1264
Gemini 1.5 ProModel Category=Closed-source Large-sized MLLM2025.1164
Gemini-1.5-Pro2025.1164
Qwen3-VL-8B-Thinking + MVPFrames=1282026.0163.8
InternVL-2.5-78BModel Category=Open-source Large-sized MLLM2025.1163.6
Qwen3-VL-8B-ThinkingFrames=1282026.0162.9
VideoChat-R1.5-7BModel Category=Open-source Medium-sized MLLM2025.1162.6
Qwen3-VL-4BAgent=disabled2026.0562.6
OryxSize=34B2024.0962.2
InternVL3.5Size=8B2026.0262.1
InternVL-3.5-8BModel Category=Open-source Medium-sized MLLM2025.1162.1
Oryx-1.5Size=32B2024.0962
Qwen3-VL-8B-Thinking + MVPFrames=642026.0162
LLaVA-VideoSize=72B2026.0261.9
Demo-ICL (SFT)Size=7B2026.0261.6
VidLaDA#S (Model Size)=8B, #F (Number of input frames)=64, Architecture=DLM Baseline2026.0161.4
LLaVA-OneVisionSize=72B2024.0961.3
LLaVA-OneVisionSize=72B2026.0261.3
LLaVA-ov-72BModel Category=Open-source Large-sized MLLM2025.1161.3
Demo-ICLSize=7B2026.0261.2
InternVL3.5-8B + MVPFrames=642026.0161
VideoXL2-8BModel Category=Open-source Medium-sized MLLM2025.1161
GPT-4V2024.0960.7
Qwen2.5-VLSize=72B2026.0260.7
Ola-Video (Base)Size=7B2026.0260.7
Qwen2.5-VL-72BFrames=7682026.0160.7
Qwen2.5-VL-72BModel Category=Open-source Large-sized MLLM2025.1160.7
InternVideo2.5-7BModel Category=Open-source Medium-sized MLLM2025.1160.6
Hour-LLaVAModel Size=7B2025.1160.4
Qwen2.5-VL#S (Model Size)=7B, #F (Number of input frames)=64, Architecture=AR Baseline, reproduced=true2026.0160.2
InternVL2.5#S (Model Size)=7B, #F (Number of input frames)=64, Architecture=AR Baseline2026.0160
InternVL-2.5-8BModel Category=Open-source Medium-sized MLLM2025.1160
VideoLLaMA3Size=7B2026.0259.8
Qwen3-VL-8B-ThinkingFrames=642026.0159.6
InternVL3.5-8B + MVPFrames=1282026.0159.5
Streamo-7BCategory=Streamo Framework, Model Size=7B, Base Model=Qwen2.5-VL-7B2025.1259.2
Streamo-7BModel Category=Streamo Framework, Parameters=7B2025.1259.2
StreamingVLM-7BCategory=Open-source Online Models, Model Size=7B2025.1259
StreamingVLM-7BModel Category=Open-source Online Models, Parameters=7B2025.1259
InternVL-3-8BModel Category=Open-source Medium-sized MLLM2025.1158.8
LLaDA-V#S (Model Size)=8B, #F (Number of input frames)=32, Architecture=DLM Baseline, reproduced=true2026.0158.6
Qwen2.5-VL-7B-Instruct + MVPFrames=1282026.0158.6
GPT-4oParams=-, Frame=322026.0258.5
LLaVA-VideoSize=7B2026.0258.2
LLaVA-Video#S (Model Size)=7B, #F (Number of input frames)=64, Architecture=AR Baseline, reproduced=true2026.0158.2
LLaVA-Video-7BModel Category=Open-source Medium-sized MLLM2025.1158.2
LongRL-7BModel Category=Open-source Medium-sized MLLM2025.1158.1
LongVILA-R1Model Size=7B2025.1158
REVISOR (Ours)Model Size=7B, Video Tokens=8K2025.1157.5
Qwen2.5-VL*Model Size=7B, Video Tokens=8K, Training=text-based reflection mechanism2025.1157.4
LongVILA-7BModel Category=Open-source Medium-sized MLLM2025.1157.1
LLaVA-OneVision#S (Model Size)=7B, #F (Number of input frames)=32, Architecture=AR Baseline, reproduced=true2026.0156.5
Streamo-2BCategory=Streamo Framework, Model Size=2B, Base Model=InternVL3-2B2025.1256.5
Qwen2.5-VL⋆Model Size=7B, Video Tokens=8K, Status=reproduction2025.1156.5
LLaVA-OneVisionModel Size=7B, Video Tokens=6K2025.1156.4
VL-RethinkerModel Size=7B, Video Tokens=8K2025.1156.4
Oryx-1.5Size=7B2024.0956.3
LLaVA-OneVisionSize=7B2026.0256.3
Jigsaw-7BFrames=1282026.0156.2
Streamo-3BCategory=Streamo Framework, Model Size=3B, Base Model=Qwen2.5-VL-3B2025.1256.2
Streamo-3BModel Category=Streamo Framework, Parameters=3B2025.1256.2
Streamo-4BCategory=Streamo Framework, Model Size=4B, Base Model=Qwen3-VL-4B2025.1256.1
Qwen2.5-VLSize=7B2026.0256
Qwen2.5-VL-7BCategory=Streamo Framework, Model Size=7B2025.1256
Qwen2.5-VL-7BModel Category=Streamo Framework, Parameters=7B2025.1256
Qwen2.5-VL-7BModel Category=Open-source Medium-sized MLLM2025.1156
Qwen2.5-VL-7B-Instruct + MVPFrames=642026.0155.9
VambaModel Size=10B2025.1155.9
InternVL3.5-8BFrames=1282026.0155.6