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

Multi-modal Reasoning on MMMU-Pro

85.6Accuracy

CoT2-Meta

Updated 15d ago

Evaluation Results

Method	Links
CoT2-Meta 2026.03		85.6	14.5
CoT2-Meta 2026.03		78.4	10.5
Vanilla ToT 2026.03		77.8	8.3
Best-of-16 2026.03		73.1	4.8
Vanilla ToT 2026.03		71.5	5.7
Greedy CoT 2026.03		68.4	-
Best-of-16 2026.03		68.2	3
Greedy CoT 2026.03		64.6	-
CoT2-Meta 2026.03		55.2	12.2
Vanilla ToT 2026.03		48.6	6.4
Best-of-16 2026.03		44.5	3.4
Greedy CoT 2026.03		41	-
Qwen2.5-VL-7B + PGPO 2026.04		39.01	-
Qwen2.5-VL-7B + VPPO 2026.04		38.75	-
Qwen2.5-VL-7B + PAPO 2026.04		38.38	-
VL-Rethinker-7B 2026.04		37.13	-
Qwen2.5-VL-7B + DAPO 2026.04		36.99	-
Qwen2.5-VL-7B + GRPO 2026.04		36.84	-
R1-ShareVL-7B 2026.04		35.1	-
NoisyRollout-7B 2026.04		34.5	-
MM-Eureka-7B 2026.04		30.3	-
Qwen2.5-VL-3B + PGPO 2026.04		29.33	-
Qwen2.5-VL-3B + VPPO 2026.04		28.83	-
Qwen2.5-VL-3B + PAPO 2026.04		28.76	-
Qwen2.5-VL-3B + GRPO 2026.04		28.02	-
Qwen2.5-VL-3B + DAPO 2026.04		27.2	-
Qwen2.5-VL-7B 2026.04		26.67	-
Qwen2.5-VL-3B 2026.04		20.95	-