List your company

Mathematical Reasoning on GSM8K (Accuracy, Token Usage, Flips)

94.54AccuracyLong base, 8,192 limit
93.354493.662293.9794.2778Jun 18, 2026

Evaluation results

MethodLinks
Long base, 8,192 limitPolicy=Long base, 8,192 limit2026.0694.5401,1570
SELECTIVE ACTIVE VERIFYPolicy=SELECTIVE ACTIVE VERIFY2026.0694.4731,3350
Base, 4,096 limitPolicy=Base, 4,096 limit2026.0693.401,1800
Always active verifyPolicy=Always active verify2026.0693.41002,9321.25