Share your thoughts, 1 month free Claude Pro on usSee more

Video Understanding on Video-MME (Accuracy)

84.8Accuracy

Gemini-2.5 Pro

Updated 22d ago

Evaluation Results

Method	Links
Gemini-2.5 Pro 2026.05		84.8
GPT-4o 2026.05		77.2
Gemini-1.5-Pro 2026.07		75
Qwen2.5-VL-72B 2026.07		73.3
GPT-4o 2026.07		71.9
QCA 2026.07		71.4
CapRL-Video-178K 2026.06		71.1
Tarsier2-7B-Video-178K 2026.06		70.9
InternVL3.5-38B 2026.07		70.9
GPT-4o-Video-178K 2026.06		70.5
ShareGPT4Video-8B-Video-178K 2026.06		70.3
CapRL-Video-178K 2026.06		70.2
LLaVA-Video-72B 2026.07		70
Qwen3-VL-30B-A3B 2026.07		69.9
Tarsier2-7B-Video-178K 2026.06		69.7
QCA 2026.07		69.5
GPT-4o-Video-178K 2026.06		68.7
ShareGPT4Video-8B-Video-178K 2026.06		68.4
VideoSeeker-8B 2026.05		68.1
Kimi-VL-16B-A3B 2026.07		67.8
Qwen3-VL-8B 2026.07		67.6
Qwen3-VL-8B 2026.05		67.4
Qwen3-VL-8B + ADPO 2026.05		66.7
InternVL3-8B + HY-Himmel 2026.05		66.3
LongVT-7B 2026.05		66.1
VideoSeeker-4B 2026.05		66.1
Qwen3-VL-8B + DAPO 2026.05		65.9
Qwen3-VL-8B + GRPO 2026.05		65.4
Qwen3-VL-4B 2026.05		65.3
AKS 2025.10		65.3
Qwen3-VL-8B + HY-Himmel 2026.05		64.9
Qwen3-VL-8B 2026.05		64.7
LLaVA-Video-7B 2026.07		64.4
InternVL3-8B baseline 2026.05		64.3
NVILA 2026.07		64
Qwen2.5-VL-7B + HY-Himmel 2026.05		63.5
Qwen3-VL-8B baseline 2026.05		62.7
EADP 2026.07		62
Video-SFT 2026.03		61.93
Hybrid-Frame Strategy 2026.03		61.52
Qwen3-VL-4B 2026.05		61.5
Video-SFT 2026.03		61.3
Apollo 2026.07		61.3
Qwen2.5-VL-7B baseline 2026.05		61.2
DivPrune 2026.07		61.2
Video-R1-7B 2026.05		61
Video-Thinker-7B 2026.05		61
Hybrid-Frame Strategy 2026.03		60.93
VideoRFT-7B 2026.05		60.9
CDPruner 2026.07		60.9
EADP 2026.07		60.4
Video-SFT 2026.03		59.9
LLaVA-OV-7B + HY-Himmel 2026.05		59.6
mPLUG-Owl3 2026.07		59.3
DivPrune 2026.07		59.2
HiPrune 2026.07		59
LLaVA-Video-7B + FrameOracle 2025.10		58.9
Video-SFT 2026.03		58.74
LLaVA-Video-7B + BOLT 2025.10		58.6
Qwen3-VL-8B 2026.05		58.3
CDPruner 2026.07		58.3
Hybrid-Frame Strategy 2026.03		58.15
LLaVA-OV-7B baseline 2026.05		58
LLaVA-Video-7B + KFC 2025.10		57.6
LLaVA-OneVision-7B + Frame-Voyager 2025.10		57.5
LLaVA-OneVision-7B + FrameOracle 2025.10		57.5
Qwen2.5-VL-7B + K-frames 2025.10		57.4
Qwen2.5-VL-7B + FrameOracle 2025.10		57.4
FP16 (Full Precision) 2026.06		57.33
Qwen2.5-VL-7B + ViaRL 2025.10		57.3
EADP 2026.07		57.3
DivPrune 2026.07		57.2
CDPruner 2026.07		56.3
LLaVA-OneVision-7B + BOLT 2025.10		56.1
HiPrune 2026.07		56
LLaVA-Video-7B 2025.10		55.9
LLaVA-OneVision-7B + KFC 2025.10		55.4
Hybrid-Frame Strategy 2026.03		54.96
QLoRA 2026.06		54.78
Video-SFT 2026.03		54.41
Qwen2.5-Omni (Ours) 2026.06		54.33
Qwen2.5-VL-7B 2025.10		54.1
LLaVA-OneVision-7B 2025.10		53.8
HiPrune 2026.07		53.8
Hybrid-Frame Strategy 2026.03		52.41
Video-SFT 2026.03		51.39
PoMQ-ViT 2026.06		49.22
QVLM 2026.06		48.41
Naive W4A4 2026.06		47.52
Video-SFT 2026.03		38.59