ResearchBenchmarksMathematical Reasoning on GSM8k (test) (Accuracy, PPL)Mathematical Reasoning on GSM8k (test) (Accuracy, PPL)Follow91.3AccuracyITI87.24488.29789.3590.403May 20, 2026Evaluation resultsMethodMethodLinksAccuracyPerplexityITIBase Model=Gemma-4-E4b-itBase Model=Gemma-4-E4b-it2026.0591.31.293MAGSBase Model=Gemma-4-E4b-itBase Model=Gemma-4-E4b-it2026.0591.31.21ASBase Model=Gemma-4-E4b-itBase Model=Gemma-4-E4b-it2026.05901.416UnsteeredBase Model=Gemma-4-E4b-itBase Model=Gemma-4-E4b-it2026.0589.91.71CDBase Model=Gemma-4-E4b-itBase Model=Gemma-4-E4b-it2026.0587.41.107