Code Generation on HumanEval (Pass@1, Token Usage)
93.29Pass@1MixReasoning
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| MixReasoningBackbone=Qwen3-8B2025.10 | 93.29 | 2,643 | |
| Qwen3-8B (Ori.)Backbone=Qwen3-8B2025.10 | 91.95 | 2,881 | |
| Concise ModeBackbone=Qwen3-8B2025.10 | 90.93 | 2,543 |