List your company

Mathematical Reasoning on GSM8K (Accuracy, Latency (s))

96.02AccuracyFlowCompile
88.719290.614692.5194.4054May 13, 2026

Evaluation results

MethodLinks
FlowCompileSelection Priority=Accuracy-priority2026.0596.0288.3
Qwen3-4BBackbone=Qwen3-4B2026.0595.6494.4
Full BaselineBackbone=Qwen3-14B2026.0595.55168.3
Qwen3-8BBackbone=Qwen3-8B2026.0595.17134.8
MaASBackbone=Qwen3-8B2026.0592.847.9
MaASBackbone=Qwen3-4B2026.0592.4230.5
FlowCompileSelection Priority=Latency-priority2026.0592.1310.9
KNN Router2026.058942.5