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

Mathematical & Geometric Reasoning on Geo3k

56.6Accuracy@8

NoisyRollout-32B

Updated 1mo ago

Evaluation Results

Method	Links
NoisyRollout-32B 2025.10		56.6
Qwen2.5-VL-32B + VPPO 2025.10		53.4
NoisyRollout-7B 2025.10		51.8
Qwen2.5-VL-32B + GRPO 2025.10		51.4
Qwen2.5-VL-32B + DAPO 2025.10		51.4
MM-Eureka-32B 2025.10		51
Qwen2.5-VL-32B 2025.10		47
Qwen2.5-VL-7B + VPPO 2025.10		46.5
PAPO-D-7B 2025.10		44.1
Qwen2.5-VL-7B + DAPO 2025.10		41.5
R1-ShareVL-7B 2025.10		41.2
VL-Rethinker-7B 2025.10		40.7
MM-Eureka-7B 2025.10		40.3
Qwen2.5-VL-7B + GRPO 2025.10		40.2
ThinkLite-7B 2025.10		37.6
Qwen2.5-VL-7B 2025.10		37.1