List your company
ResearchFormerly WizWand

Long Video Understanding on LongVideoBench (Accuracy)

80.3AccuracySeed2.0 Pro
40.2650.65561.0571.445Mar 6, 2026Mar 25, 2026Apr 13, 2026May 3, 2026May 22, 2026Jun 10, 2026Jun 30, 2026
Updated 4mo ago

Evaluation results

MethodLinks
Seed2.0 Pro2026.0680.3
Seed1.82026.0677.4
Seed2.0 Lite2026.0677.3
Gemini-3-Pro2026.0676.7
Seed2.0 Mini2026.0674.8
Gemini-3-Flash2026.0674.5
Penguin-VLModel Size=8B2026.0367
GPT-4o2026.0466.7
GPT-4oSize=-2026.0466.7
GPT-4oPerception budget constraint=false, Perc. budget=-2026.0566.7
Eagle2.5-8BSize=8B2026.0466.4
VideoChat-Flash-7BSize=7B2026.0464.7
Gemini-1.5-ProSize=-2026.0464
Qwen3-VLArchitecture Type=Modular, Parameter Scale=8B, Training Stage=Instruct2026.0563.6
NEO-ovArchitecture Type=Native, Parameter Scale=8B, Training Stage=Instruct2026.0563.5
RLER2026.0463
Qwen3-VLModel Size=8B2026.0362.6
InternVL3-9BSize=9B2026.0462.5
InternVL-3.5Model Size=8B2026.0362.1
InternVL3.5Architecture Type=Modular, Parameter Scale=8B, Training Stage=Instruct2026.0562.1
Video-ZeroBackbone=InternVL3.5-4B, Checkpoint=v32026.0561.2
Video-ZeroBackbone=InternVL3.5-4B, Checkpoint=v12026.0561
Video-ZeroBackbone=InternVL3.5-4B, Checkpoint=v52026.0560.9
VisPlayBackbone=Qwen3-VL-8B-Instruct, Iteration=22026.0560.7
Video-ZeroBackbone=InternVL3.5-4B, Checkpoint=v42026.0560.7
V-ZeroBackbone=Qwen3-VL-8B-Instruct, Iteration=32026.0560.6
Video-ZeroBackbone=Qwen3-VL-8B-Instruct, Iteration=12026.0560.6
Video-Zero (Peak)Backbone=Qwen3-VL-8B-Instruct2026.0560.6
VisPlayBackbone=Qwen3-VL-8B-Instruct, Iteration=42026.0560.5
VisPlayBackbone=Qwen3-VL-8B-Instruct2026.0560.5
VisPlayBackbone=Qwen3-VL-8B-Instruct, Iteration=52026.0560.4
V-ZeroBackbone=Qwen3-VL-8B-Instruct, Iteration=12026.0560.4
Video-ZeroBackbone=Qwen3-VL-8B-Instruct, Iteration=32026.0560.4
Video-ZeroBackbone=Qwen3-VL-8B-Instruct, Iteration=42026.0560.4
Video-Zero (Iter 4)Backbone=Qwen3-VL-8B-Instruct2026.0560.4
Video-ZeroBackbone=InternVL3.5-4B, Checkpoint=v22026.0560.4
V-ZeroBackbone=Qwen3-VL-8B-Instruct, Iteration=52026.0560.3
InternVL3.5-4BCheckpoint=Base2026.0560.3
Video-ZeroBackbone=Qwen3-VL-8B-Instruct, Iteration=22026.0560.2
Video-ZeroBackbone=Qwen3-VL-8B-Instruct, Iteration=52026.0560.2
V-ZeroBackbone=Qwen3-VL-8B-Instruct, Iteration=42026.0560.1
V-ZeroBackbone=Qwen3-VL-8B-Instruct2026.0560.1
VisPlayBackbone=Qwen3-VL-8B-Instruct, Iteration=32026.0560
mPLUG-Owl3-8B2026.0459.8
VideoLLaMA3-7BPerception budget constraint=false, Perc. budget=-2026.0559.8
Base ModelBackbone=Qwen3-VL-8B-Instruct2026.0559.8
VisPlayBackbone=Qwen3-VL-8B-Instruct, Iteration=12026.0559.8
V-ZeroBackbone=Qwen3-VL-8B-Instruct, Iteration=22026.0559.8
Base ModelBackbone=Qwen3-VL-8B-Instruct2026.0559.8
VideoLLaMA3Architecture Type=Modular, Parameter Scale=8B, Training Stage=Instruct2026.0559.8
XCompSize=2B2026.0459.7
VisPlayBackbone=Qwen3-VL-4B-Instruct, Iteration=22026.0559.6
V-ZeroBackbone=Qwen3-VL-4B-Instruct, Iteration=32026.0559.6
Video-ZeroBackbone=Qwen3-VL-4B-Instruct, Iteration=52026.0559.6
Video-Zero (Peak)Backbone=Qwen3-VL-4B-Instruct2026.0559.6
LLaVA-Video-7B2026.0459.5
V-ZeroBackbone=Qwen3-VL-4B-Instruct, Iteration=42026.0559.5
V-ZeroBackbone=Qwen3-VL-4B-Instruct2026.0559.5
Video-ZeroBackbone=Qwen3-VL-4B-Instruct, Iteration=22026.0559.4
VisPlayBackbone=Qwen3-VL-4B-Instruct, Iteration=52026.0559.3
VisPlayBackbone=Qwen3-VL-4B-Instruct, Iteration=42026.0559.2
V-ZeroBackbone=Qwen3-VL-4B-Instruct, Iteration=22026.0559.2
V-ZeroBackbone=Qwen3-VL-4B-Instruct, Iteration=52026.0559.2
VisPlayBackbone=Qwen3-VL-4B-Instruct2026.0559.2
GPT-4VSize=-2026.0459.1
Video-ZeroBackbone=Qwen3-VL-4B-Instruct, Iteration=32026.0559.1
Video-ZeroBackbone=Qwen3-VL-4B-Instruct, Iteration=42026.0559.1
Video-Zero (Iter 4)Backbone=Qwen3-VL-4B-Instruct2026.0559.1
VisPlayBackbone=Qwen3-VL-4B-Instruct, Iteration=12026.0558.9
V-ZeroBackbone=Qwen3-VL-4B-Instruct, Iteration=12026.0558.6
VideoChat-Flash-2BSize=2B2026.0458.3
Video-ZeroBackbone=Qwen3-VL-4B-Instruct, Iteration=12026.0558.2
LLaVA-VideoArchitecture Type=Modular, Parameter Scale=8B, Training Stage=Instruct2026.0558.2
VisPlayBackbone=Qwen3-VL-4B-Instruct, Iteration=32026.0558
Base ModelBackbone=Qwen3-VL-4B-Instruct2026.0557.7
Base ModelBackbone=Qwen3-VL-4B-Instruct2026.0557.7
MACF (LLaVA-OV1.5-8B)Perception budget constraint=true, Perc. budget=16*224*224, Backbone=LLaVA-OneVision1.5-8B2026.0557.6
InternVL3.5Architecture Type=Modular, Parameter Scale=2B, Training Stage=Instruct2026.0557.4
VideoLLaMA3Architecture Type=Modular, Parameter Scale=2B, Training Stage=Instruct2026.0557.1
MACF (Qwen3-VL-8B)Perception budget constraint=true, Perc. budget=16*224*224, Backbone=Qwen3-VL-8B2026.0556.8
NEO-ovArchitecture Type=Native, Parameter Scale=2B, Training Stage=Instruct2026.0556.8
Keye1.5-VL-8BPerception budget constraint=true, Perc. budget=16*224*2242026.0556.4
TimeMarkerSize=8B2026.0456.3
Qwen2.5-VL-7B2026.0456
Qwen3-VL-30BPerception budget constraint=true, Perc. budget=16*224*2242026.0555.9
Qwen3-VLArchitecture Type=Modular, Parameter Scale=2B, Training Stage=Instruct2026.0555.6
InternVL3-2BSize=2B2026.0455.4
KangarooSize=8B2026.0454.8
Kangaroo-8BPerception budget constraint=false, Perc. budget=-2026.0554.8
LLaVA-OneVision1.5-8BPerception budget constraint=true, Perc. budget=16*224*2242026.0554.4
MACF (Qwen2.5-VL-7B)Perception budget constraint=true, Perc. budget=16*224*224, Backbone=Qwen2.5-VL-7B2026.0552.5
mPLUG-Owl3Size=7B2026.0452.1
Qwen2.5-VL-72BPerception budget constraint=true, Perc. budget=16*224*2242026.0551
Qwen3-VL-8BPerception budget constraint=true, Perc. budget=16*224*2242026.0550.7
LLaVA-Next-Video-34BPerception budget constraint=false, Perc. budget=-2026.0550.5
LatentMASPerception budget constraint=true, Perc. budget=16*224*224, VLM engine=Qwen3-VL-8B2026.0548.5
Qwen2.5-VL-7BPerception budget constraint=true, Perc. budget=16*224*2242026.0548.1
InternVL2.5-VL-8BPerception budget constraint=true, Perc. budget=16*224*2242026.0543.4
Qwen2.5VL-3BSize=3B2026.0443.3
ShareGPT4Video-8BPerception budget constraint=false, Perc. budget=-2026.0541.8