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

Multimodal mathematical reasoning on MathVista (test)

74.2Accuracy

Qwen-VL-Max

Updated 18d ago

Evaluation Results

Method	Links
Qwen-VL-Max 2026.01		74.2	-
GPRO-7B 2026.01		74.2	115.3
FAST-7B 2026.01		73.8	120.7
Vision-R1 2026.01		73.5	-
GSPO + KAWHI (Qwen3) 2026.03		72.1	55.54
GRPO + KAWHI (Qwen3) 2026.03		71.8	51.98
DAPO + KAWHI (Qwen3) 2026.03		71.21	53.06
GRPO (Qwen3) 2026.03		71.07	51.17
MM-R1 2026.01		71	185.6
GSPO (Qwen3) 2026.03		70.5	54.14
Qwen3-VL-4B-Instruct Base 2026.03		70.43	50.53
OpenVLThinker 2026.01		70.2	305.7
GSPO 2026.03		69.42	51.4
DAPO (Qwen3) 2026.03		69.4	51.8
GSPO + KAWHI 2026.03		69.37	52.94
GRPO + KAWHI 2026.03		69.3	51
DAPO 2026.03		69.2	50.79
DAPO + KAWHI 2026.03		69.08	51.85
VPPO 2026.03		68.42	50.29
GRPO 2026.03		68.37	50.18
Qwen2.5-VL-7B 2026.01		68.2	189.1
Claude-3.5 Sonnet 2026.01		67.7	-
FT-RL 2026.03		67.68	49.37
Qwen2.5-VL-7B-Instruct Base 2026.03		67.47	47.68
Step-GRPO 2026.03		66.87	47.82
GPRO-3B 2026.01		66.8	145.2
FAST-3B 2026.01		66.2	158.7
R1-OneVision 2026.01		64.1	402.5
GPT-4o 2026.01		63.8	-
LMM-R1 2026.01		63.2	245
Qwen2.5-VL-3B 2026.01		62.3	212.9
Mulberry 2026.01		62.1	275
Curr-ReFT 2026.01		61.9	95.9
Qwen2-VL-7B 2026.01		58.2	265.9