Mathematical Reasoning on MATH 500 (Accuracy, Latency)
96.77AccuracyQwen3-8B
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Qwen3-8BBackbone=Qwen3-8B2026.05 | 96.77 | 352.3 | |
| Full BaselineBackbone=Qwen3-14B2026.05 | 96.52 | 355.5 | |
| FlowCompileSelection Priority=Accuracy-priority2026.05 | 95.27 | 185.5 | |
| Qwen3-4BBackbone=Qwen3-4B2026.05 | 93.78 | 187.4 | |
| FlowCompileSelection Priority=Latency-priority2026.05 | 85.32 | 31.9 | |
| KNN Router2026.05 | 78.61 | 82 | |
| MaASBackbone=Qwen3-4B2026.05 | 71.82 | 55 | |
| MaASBackbone=Qwen3-8B2026.05 | 70.4 | 87.1 |