List your company

Commonsense Reasoning on CommonsenseQA (Accuracy, Extra Calls, Token Usage, Flips)

85.18AccuracySC sampled-rollout oracle
71.805675.277878.7582.2222Jun 18, 2026

Evaluation results

MethodLinks
SC sampled-rollout oraclePolicy=SC sampled-rollout oracle2026.0685.18———
Sampled-rollout oracle, SC samplesPolicy=Sampled-rollout oracle, SC samples2026.0685.18———
Expected-action oracle, SC samplesPolicy=Expected-action oracle, SC samples2026.0680.98———
Self-Consistency@5Policy=Self-Consistency@52026.0678.38511,3431.56
Self-Consistency@5Policy=Self-Consistency@52026.0678.38—11,3431.56
Expected-action oracle, active verifyPolicy=Expected-action oracle, active verify2026.0678.26———
Base, 4,096 limitPolicy=Base, 4,096 limit2026.0676.4912,2340
Accept basePolicy=Accept base2026.0676.49—00
Always active verifyPolicy=Always active verify2026.0672.3224,7945.94
Always active verifyPolicy=Always active verify2026.0672.32—2,5615.94