Code Generation on LiveCodeBench (AUCOAA)
94.3AUCOAABase model
Evaluation Results
| Method | Links | |
|---|---|---|
| Base modelBackbone=Qwen3-8B2026.01 | 94.3 | |
| Hard-Length 16kBackbone=Qwen3-8B2026.01 | 94.1 | |
| SFTBackbone=Qwen3-8B2026.01 | 93.9 | |
| TWYNBackbone=Qwen3-8B2026.01 | 93.9 | |
| Format-Adaptive-AnswerBackbone=Qwen3-8B2026.01 | 93.6 | |
| Hard-Length 8kBackbone=Qwen3-8B2026.01 | 93.1 | |
| Adaptive-AnswerBackbone=Qwen3-8B2026.01 | 93.1 | |
| Soft-LengthBackbone=Qwen3-8B2026.01 | 93 | |
| Hard-Length 8k → 4kBackbone=Qwen3-8B2026.01 | 92.7 | |
| Normalized-LengthBackbone=Qwen3-8B2026.01 | 90 | |
| No-ThinkingBackbone=Qwen3-8B2026.01 | 41 |