Code Generation on HumanEval (held-out)
95.1AccuracySC-Route
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| SC-RouteBackbone=Qwen3-32B2026.06 | 95.1 | — | — | |
| DARTBackbone=Qwen3-32B2026.06 | 95.1 | 63 | 76.8 | |
| NTBackbone=Qwen3-32B2026.06 | 79.9 | — | — | |
| SC-RouteBackbone=Qwen3-8B2026.06 | 78.7 | — | — | |
| DARTBackbone=Qwen3-8B2026.06 | 78.7 | 55 | 54.9 | |
| SC-RouteBackbone=Qwen3-14B2026.06 | 78.7 | — | — | |
| DARTBackbone=Qwen3-14B2026.06 | 78.7 | 60 | 68.9 | |
| ATBackbone=Qwen3-32B2026.06 | 72.6 | — | — | |
| NTBackbone=Qwen3-14B2026.06 | 71.3 | — | — | |
| ATBackbone=Qwen3-14B2026.06 | 66.5 | — | — | |
| NTBackbone=Qwen3-8B2026.06 | 60.4 | — | — | |
| ATBackbone=Qwen3-8B2026.06 | 59.1 | — | — |