List your company

Scientific Reasoning on Scientific Methods Reasoning Cross-Domain (test)

49.08AccuracyReason-SFT-DrGRPO
1.728814.021926.31538.6081Apr 6, 2026

Evaluation results

MethodLinks
Reason-SFT-DrGRPOModel=Qwen32026.0449.08
Reason-SFT-DAPOModel=Qwen32026.0448.37
Reason-DAPOModel=Qwen32026.0445.96
Reasoning (high)Model=GPT-52026.0445.96
Direct-SFTModel=Qwen32026.0445.67
Reasoning (med)Model=GPT-52026.0445.25
Reasoning (low)Model=GPT-52026.0442.84
Synthetic-Reason-SFT-DAPOModel=Qwen32026.0441.1
Direct-SFTModel=Llama3.12026.0431.2
Reason-SFTModel=Qwen32026.0429.31
Base (Reasoning)Model=Llama3.12026.0418.22
Base (Reasoning)Model=Qwen32026.0412.62
BaseModel=Llama3.12026.043.83
BaseModel=Qwen32026.043.55