List your company
ResearchFormerly WizWand

Video Question Answering on Video-MME (Long, No Subtitles)

81.8AccuracyGPT-5
31.25644.37857.570.622Mar 18, 2026Apr 4, 2026Apr 22, 2026May 10, 2026May 27, 2026Jun 14, 2026Jul 2, 2026
Updated 4mo ago

Evaluation results

MethodLinks
GPT-5Active=Prop.2026.0381.8
VideoARM (GPT-o3+GPT-4o)Active=Prop.2026.0381.2
Claude-Opus-4.5Active=Prop.2026.0377.6
Video-RLM (Gemini-3-Flash)Active=Prop.2026.0376.2
GLM-4.5V-106BActive=12B2026.0374.6
InternVL3.5-241BActive=28B2026.0372.9
GPT-4oActive=Prop.2026.0365.3
LLM over CaptionsActive=Prop.+3B2026.0364.2
Qwen3.5 (uniform, 160 fr.)Active=3B2026.0363.8
Qwen2-VL + ReQuestLLM Size=8B, #Frames=≤ 512, zero-shot transfer=true2026.0762.4
MR.Video (Gemini+GPT-4o)Active=Prop.2026.0361.8
Qwen2-VLLLM Size=8B, #Frames=512, reproduced=true2026.0761.2
LongVULLM Size=7B, #Frames=1fps2026.0759.5
LLaVA-Video + ReQuestLLM Size=7B, #Frames=322026.0755.8
VideoARM (Qwen3-VL-235B)Active=22B2026.0354.9
NVILALLM Size=8B, #Frames=10242026.0754.8
VideoTreeLLM Size=–, #Frames=–2026.0754.2
LLaVa-OneVision + ReQuestLLM Size=7B, #Frames=32, zero-shot transfer=true2026.0752.3
LLaVA-VideoLLM Size=7B, #Frames=32, reproduced=true2026.0752.2
Video-RLM (Qwen3.5-3B)Active=3B2026.0350.4
LLaVa-OneVisionLLM Size=7B, #Frames=32, reproduced=true2026.0749.2
Frame-VoyagerLLM Size=7B, #Frames=82026.0748.9
LLoViLLM Size=–, #Frames=–2026.0748.8
VideoLLaMA2LLM Size=7B, #Frames=162026.0742.1
TimeSuiteLLM Size=7B, #Frames=1282026.0741.9
Video-LLaVALLM Size=7B, #Frames=82026.0736.2
Chat-UniVi-V1.5LLM Size=7B, #Frames=642026.0735.8
ShareGPT4VideoLLM Size=8B, #Frames=162026.0735
VideoChat2LLM Size=7B, #Frames=162026.0733.2