List your company

Mathematical Reasoning on MATH (Correction Uplift %)

48.87Correction UpliftROSA
2.257214.358626.4638.5614Sep 27, 2025

Evaluation results

MethodLinks
ROSAModel=Qwen3-0.6B2025.0948.87
ROSAModel=Qwen2.5-7B-Instruct2025.0941.53
ROSAModel=Qwen3-8B2025.0940.42
ROSAModel=Qwen2.5-0.5B-Instruct2025.0925.48
BaselineModel=Qwen3-8B2025.0923
BaselineModel=Qwen3-0.6B2025.0917.4
BaselineModel=Qwen2.5-7B-Instruct2025.0912.54
BaselineModel=Qwen2.5-0.5B-Instruct2025.096.88
ROSAModel=DeepSeek-R1-Distill-Llama-8B2025.096.3
BaselineModel=DeepSeek-R1-Distill-Llama-8B2025.094.05