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

Video Understanding on VideoMMMU

68.64Accuracy

Vanilla MoE

Updated 18d ago

Evaluation Results

Method
Vanilla MoE 2026.04	68.64	-	-	-	-	-
MACS (Ours) 2026.04	68.49	-	-	-	-	-
MACS (w/o Expanded) 2026.04	67.92	-	-	-	-	-
VideoSearcher-8B 2026.07	66.33	-	-	-	-	-
Qwen3-VL-8B-Instruct 2026.07	65.67	-	-	-	-	-
VideoSearcher-4B 2026.07	65.33	-	-	-	-	-
Vanilla MoE 2026.04	65.24	-	-	-	-	-
MACS (Ours) 2026.04	65.03	-	-	-	-	-
MACS (w/o Expanded) 2026.04	64.58	-	-	-	-	-
CAI-MoE (Expanded) 2026.04	64.28	-	-	-	-	-
CAI-MoE (Token Drop) 2026.04	62.17	-	-	-	-	-
GPT-4o 2025.12	61.2	-	-	-	-	-
GPT-4o 2026.05	61.2	-	-	-	-	-
CAI-MoE (Expanded) 2026.04	61.17	-	-	-	-	-
Qwen3-VL-4B-Instruct 2026.07	60.78	-	-	-	-	-
CAI-MoE (Token Drop) 2026.04	58.63	-	-	-	-	-
Vanilla MoE 2026.04	57.58	-	-	-	-	-
MACS (Ours) 2026.04	57.49	-	-	-	-	-
MACS (w/o Expanded) 2026.04	57.12	-	-	-	-	-
AdaTooler-V-7B 2025.12	56.8	-	-	-	-	-
AdaTooler-V-7B 2025.12	55.1	-	-	-	-	-
CAI-MoE (Expanded) 2026.04	54.63	-	-	-	-	-
AdaTooler-V-7B 2025.12	54.6	-	-	-	-	-
Gemini 1.5 Pro 2025.12	53.9	-	-	-	-	-
CAI-MoE (Token Drop) 2026.04	52.87	-	-	-	-	-
Video-RTS-7B 2026.05	52.7	-	-	-	-	-
Video-R1 2025.12	52.4	-	-	-	-	-
Video-R1 2026.07	52.4	-	-	-	-	-
Video-XL 2025.12	52.3	-	-	-	-	-
Open-o3 Video 2025.12	52.3	-	-	-	-	-
ViSS-R1-7B 2026.05	51.7	-	-	-	-	-
Qwen2.5-VL-7B + SFT + SAGE 2026.05	51.7	-	-	-	-	-
VideoChat-R1.5 2026.07	51.4	-	-	-	-	-
Qwen2.5-VL-7B-Instruct 2025.12	51.3	-	-	-	-	-
VideoRFT-7B 2026.05	51.1	-	-	-	-	-
VideoRFT 2026.07	51.1	-	-	-	-	-
Qwen2.5-VL-7B + SAGE 2026.05	50.6	-	-	-	-	-
VideoCoM 2026.07	50.2	-	-	-	-	-
Video-R1-7B 2026.05	49.8	-	-	-	-	-
VidGround 2026.04	49.4	-	-	-	-	-
Qwen2.5-VL-7B-Instruct 2025.12	49.1	-	-	-	-	-
TW-GRPO 2026.04	48.6	-	-	-	-	-
VidGround 2026.04	48.3	-	-	-	-	-
TW-GRPO 2026.04	48.3	-	-	-	-	-
TW-GRPO 2026.04	47.9	-	-	-	-	-
Qwen2.5-VL-7B-SFT 2026.04	47.8	-	-	-	-	-
Video-RTS 2026.04	47.7	-	-	-	-	-
Qwen2.5-VL-7B-SFT 2026.04	47.6	-	-	-	-	-
Qwen2.5-VL-7B-Instruct 2025.12	47.4	-	-	-	-	-
VidGround 2026.04	47.4	-	-	-	-	-
Video-RTS 2026.04	47.1	-	-	-	-	-
Qwen2.5-VL-7B 2026.04	46.6	-	-	-	-	-
Video-RTS 2026.04	46.4	-	-	-	-	-
Qwen2.5-VL-7B 2026.05	46.1	-	-	-	-	-
Qwen2.5-VL-7B 2026.04	45.4	-	-	-	-	-
Video-R1 2026.04	45.4	-	-	-	-	-
Video-R1 2026.04	45.4	-	-	-	-	-
Qwen2.5-VL-7B 2026.04	45	-	-	-	-	-
Video-R1 2026.04	44.7	-	-	-	-	-
Qwen2.5-VL-7B + SFT 2026.05	43.4	-	-	-	-	-
Qwen2.5-VL-7B-SFT 2026.04	43.1	-	-	-	-	-
LongVILA-R1-7B 2026.04	41.2	-	-	-	-	-
LongVILA-R1-7B 2026.04	40.7	-	-	-	-	-
LongVILA-R1-7B 2026.04	38.8	-	-	-	-	-
VILA-1.5-40B 2026.05	34	-	-	-	-	-
LLaVA-OneVision-7B 2026.05	33.8	-	-	-	-	-
LongVA-7B 2026.05	23.9	-	-	-	-	-
Gemini-1.5-Pro 2026.02	-	49.3	53.3	59	53.8	-
GPT-4o 2026.02	-	55.6	62	66	61.2	-
VILA-1.5-40B 2026.02	-	32.6	30.6	38.6	34	-
VideoRFT-7B 2026.02	-	-	-	-	51.1	-
LongVT-7B-RFT 2026.02	-	35.7	43.7	56.7	45.3	-
VideoChat-R1-7B 2026.02	-	40.2	45.6	53	46.8	-
Video-R1-7B 2026.02	-	44.2	46	67.3	52.4	-
Qwen2.5-VL-7B 2026.02	-	35.9	36.1	57.6	43.2	-
VideoTemp-o3-7B-SFT 2026.02	-	39.3	46.4	60.4	48.7	-
VideoTemp-o3-7B-RL 2026.02	-	43	47.8	69	53.2	-
GPT-4o 2026.03	-	-	-	-	61.2	66
EgoR1 2026.03	-	-	-	-	34.4	37.6
LongVT-RL-7B 2026.03	-	-	-	-	42.6	50
Qwen2.5-VL-7B 2026.03	-	-	-	-	51.2	64.7
VideoRFT-7B 2026.03	-	-	-	-	51.1	66
VideoR1-7B 2026.03	-	-	-	-	52.4	65.3
Open-o3-video-7B 2026.03	-	-	-	-	52.3	68
VISIONCOACH-7B 2026.03	-	-	-	-	54.4	70.3
Gemini 2.5 Pro 2026.03	-	-	-	-	83.6	-
Gemini-3-Pro 2026.03	-	-	-	-	87.6	-
Seed-1.5-VL 2026.03	-	-	-	-	81.4	-
Seed1.8 2026.03	-	-	-	-	82.7	-
Video-R1 2026.06	-	-	-	-	52.4	-
Qwen3-VL-8B 2026.06	-	-	-	-	57	-
DELTAVID-Qwen3-VL-8B 2026.06	-	-	-	-	64.56	-