List your company

Mathematical Reasoning on MATH level 3

56AccuracyGradient Shift
36.2441.3746.551.63Jan 31, 2026Feb 5, 2026Feb 10, 2026Feb 15, 2026Feb 20, 2026Feb 25, 2026Mar 2, 2026

Evaluation results

MethodLinks
Gradient Shiftvariant=mult2026.015696-0.12—
Gradient Shiftvariant=hybrid2026.015592-0.15—
Benchmark LLM (BenchAcc)2026.0152———
ME-ICPOBackbone=Qwen2.5-Math-7B, ME-ICPO=true2026.0351.9——58.87
Weak-to-Strongalpha=1.0, k=402026.015180-0.23—
Weak-to-Strongalpha=1.5, k=402026.015080-0.2—
ME-ICPOBackbone=Qwen2.5-Math-1.5B, ME-ICPO=true2026.0347.81——52.08
Weak-to-Strongalpha=2.0, k=402026.014677-0.15—
Qwen2.5-Math-7BME-ICPO=false2026.0343.71——49.88
Qwen2.5-Math-1.5BME-ICPO=false2026.0339.77——45.42
Gradient Shiftvariant=add2026.013866-0.11—
Helper LLM2026.0137———