List your company

Mathematical Reasoning on MMLU-Pro Math

89.79AccuracyMAC
30.312445.753761.19576.6363Jul 25, 2025Aug 29, 2025Oct 4, 2025Nov 9, 2025Dec 15, 2025Jan 20, 2026Feb 25, 2026

Evaluation results

MethodLinks
MAC2025.0789.7991.5293.479988.77
Debate2025.0786.7588.1989.9298.6985.38
QWQParameters=32B2025.0785.1289.2493.8198.5984.06
Deepseek-R1Parameters=32B2025.0784.988.4592.498.5683.69
OpenthinkerParameters=32B2025.0784.8388.3492.3198.5583.6
Phi4Parameters=14B2025.0779.4280.982.7997.9577.28
Qwen2.5-14B (Teacher)Model Scale=14B2026.0277.6————
Qwen2.5-14BRole=Teacher2026.0277.6————
Qwen2.5Parameters=32B2025.0776.6877.8879.3797.6774.2
Qwen2.5Parameters=14B2025.0773.9575.0776.697.3971.17
TSD-KDselection=Adaptive c%2026.0238.5————
TSD-KDDistillation Setup=Qwen2.5 (14B -> 1.5B)2026.0236.9————
MiniLLMDistillation Setup=Qwen2.5 (14B -> 1.5B)2026.0236.2————
MiniLLM2026.0236.2————
GKDbeta=0.9, Distillation Setup=Qwen2.5 (14B -> 1.5B)2026.0235.7————
GKDbeta=0.92026.0235.7————
Supervised-KDDistillation Setup=Qwen2.5 (14B -> 1.5B)2026.0235.3————
Supervised-KD2026.0235.3————
DistiLLMDistillation Setup=Qwen2.5 (14B -> 1.5B)2026.0235.2————
DistiLLM2026.0235.2————
Speculative KDDistillation Setup=Qwen2.5 (14B -> 1.5B)2026.0234.8————
Speculative KD2026.0234.8————
Qwen2.5-1.5B (Student)Model Scale=1.5B2026.0234.4————
Qwen2.5-1.5BRole=Student2026.0234.4————
Sequence-Level KDDistillation Setup=Qwen2.5 (14B -> 1.5B)2026.0232.6————
Sequence-Level KD2026.0232.6————