List your company

Mathematical Reasoning on MATH 500 (Accuracy, TNFT, Tokens, Delay)

64AccuracyOurs
47.56851.83456.160.366May 12, 2026

Evaluation results

MethodLinks
OursBackbone=Qwen3-4B2026.05640742.2626.51
VanillaBackbone=Qwen3-4B2026.0560.8103.161,363.9545.82
BaseBackbone=Qwen3-4B2026.0560130.13,678.51126.9
OursBackbone=Qwen3-1.7B2026.0551.6080322.94
BaseBackbone=Qwen3-1.7B2026.0548.4130.13,229.1688.15
VanillaBackbone=Qwen3-1.7B2026.0548.2103.161,612.7643.14