List your company

Mathematical Reasoning on GSM8K (Accuracy, Performance Gain %)

96.51AccuracyQwen3-14B
88.938890.904492.8794.8356Jun 4, 2026

Evaluation results

MethodLinks
Qwen3-14BSetting=Recur2026.0696.510.39
Qwen3-32BSetting=Base2026.0696.44—
Qwen3-8BSetting=Recur2026.0696.360.47
Qwen3-32BSetting=Recur2026.0696.36-0.08
Qwen3-14BSetting=Base2026.0696.13—
Qwen3-8BSetting=Base2026.0695.91—
LLaMA-70BSetting=Recur2026.0695.731.5
Qwen3-4BSetting=Recur2026.0695.530.13
Qwen3-4BSetting=Base2026.0695.41—
LLaMA-70BSetting=Base2026.0694.31—
LLaMA-8BSetting=Recur2026.0690.781.73
Qwen3-1.7BSetting=Base2026.0690.03—
Qwen3-1.7BSetting=Recur2026.0690.030
LLaMA-8BSetting=Base2026.0689.23—