List your company

Mathematical Reasoning on Headline Benchmarks Paper-specific original

25Delta Accuracy (pp)SDFT
1.2887.44413.619.756May 7, 2026

Evaluation results

MethodLinks
SDFTThinking-enabled=false, Math-reasoning=false, Model Scale=7-8B2026.0525—
SDPOThinking-enabled=false, Math-reasoning=false, Model Scale=7-8B2026.052468.8
OPSDThinking-enabled=true, Math-reasoning=true, Model Scale=7-8B2026.052.2—