ResearchBenchmarksMathematical Reasoning on AIME 2024 (Mean@16, Acc)Follow79.17Mean@16 AccuracyME-ICPO18.27834.086549.89565.7035Mar 2, 2026Evaluation ResultsMethodMethodLinksMean@16 AccuracyAccuracyME-ICPOModel=Gemini-2.5-ProModel=Gemini-2.5-Pro2026.0379.1780ME-ICPOModel=Gemini-2.5-FlashModel=Gemini-2.5-Flash2026.0376.4676.47BaseModel=Gemini-2.5-ProModel=Gemini-2.5-Pro2026.0358.5456.6ME-ICPOModel=Qwen3-4B-InstructModel=Qwen3-4B-Instruct2026.0357.7157.67BaseModel=Gemini-2.5-FlashModel=Gemini-2.5-Flash2026.0335.2135.42BaseModel=Qwen3-4B-InstructModel=Qwen3-4B-Instruct2026.0320.6220.59