ResearchBenchmarksLogical Reasoning Verification on SAIR official benchmark (hard2)Follow48Official AccuracyAN45c40.7242.6144.546.39Apr 20, 2026Evaluation ResultsMethodMethodLinksOfficial AccuracyOfficial F1 ScoreDelta vs Baseline (pp)AN45cModel=GPT-OSS 120B, n=20Model=GPT-OSS 120B, n=202026.044861.28.3AN38Model=GPT-OSS 120B, n=20Model=GPT-OSS 120B, n=202026.044147.815.3