Code generation on LiveCodeBench (Score, Avg)
57ScoreRL_RACES
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| RL_RACESBackbone=Qwen3-14B2026.06 | 57 | 61.1 | |
| RL_individualBackbone=Qwen3-14B2026.06 | 56.3 | 60.1 | |
| BaseBackbone=Qwen3-14B2026.06 | 55 | 58.8 | |
| RL_RACESBackbone=DeepSeek-R1-Distill-Qwen-14B2026.06 | 48.8 | 51.3 | |
| BaseBackbone=DeepSeek-R1-Distill-Qwen-14B2026.06 | 47.2 | 48.2 | |
| RL_individualBackbone=DeepSeek-R1-Distill-Qwen-14B2026.06 | 46.9 | 48.8 |