List your company

Math Reasoning on Math500, AIME24, AIME25, Minerva, and OlympiadBench (test)

92.2Math500 AccuracyOpenThinker3-7B
67.2473.7280.286.68Mar 26, 2026

Evaluation results

MethodLinks
OpenThinker3-7B2026.0392.253.3403955.656
teacher top-K local support matchingmasking=false2026.038233.316.732.74441.7
teacher top-K local support matchingmasking=true2026.03792016.734.642.538.6
Sampled-token OPDmasking=true2026.03762013.333.540.436.6
Sampled-token OPDmasking=false2026.0374.813.313.332.140.534.8
Qwen2.5-7B-It2026.0368.213.3026.532.928.2