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

Multimodal Math Reasoning on MMK12

80.83Accuracy

Qwen2.5-VL-7B + PGPO

Updated 16d ago

Evaluation Results

Method	Links
Qwen2.5-VL-7B + PGPO 2026.04		80.83
Qwen2.5-VL-7B + PAPO 2026.04		80.58
Qwen2.5-VL-7B + VPPO 2026.04		80.11
Qwen2.5-VL-7B + DAPO 2026.04		78.74
Qwen2.5-VL-7B + GRPO 2026.04		76.45
R1-ShareVL-7B 2026.04		70.94
VL-Rethinker-7B 2026.04		68.3
MM-Eureka-7B 2026.04		67.68
Qwen2.5-VL-3B + PGPO 2026.04		64.74
Qwen2.5-VL-3B + VPPO 2026.04		63.76
Qwen2.5-VL-3B + PAPO 2026.04		63.54
Qwen2.5-VL-3B + DAPO 2026.04		63.36
Qwen2.5-VL-3B + GRPO 2026.04		62.95
AVATAR 2025.08		57.8
AVATAR 2025.08		57
Qwen2.5-VL-7B + GRPO 2025.08		56.3
AVATAR 2025.08		56.2
AVATAR 2025.08		55.6
OpenVLThinker-7B 2025.08		53.5
VLAAThinker-VL-7B 2025.08		51.7
NoisyRollout-7B 2026.04		50
Qwen2.5-VL-7B 2025.08		47.9
Qwen2.5-VL-7B 2026.04		43.04
Qwen2.5-VL-3B 2026.04		37.37