Code Reasoning on LiveCodeBench (Avg@16, Pass@16)
43.2Avg@16ResRL
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ResRLBackbone=Qwen3-4B2026.05 | 43.2 | 59.9 | |
| FlowRLBackbone=Qwen3-4B2026.05 | 42.4 | 58.7 | |
| DAPOBackbone=Qwen3-4B2026.05 | 41 | 52.3 | |
| GRPOBackbone=Qwen3-4B2026.05 | 39.5 | 55.1 | |
| CaSPModel=DeepSeek-R1-Distill-Qwen-7B2025.10 | 36.02 | 54.48 | |
| GRPOModel=DeepSeek-R1-Distill-Qwen-7B2025.10 | 33.92 | 53.41 | |
| LAD2026.02 | 33.51 | 51.97 | |
| FlowRL2026.02 | 33.24 | 51.61 | |
| NSRBackbone=Qwen3-4B2026.05 | 32.8 | 52.3 | |
| EntAdv2026.02 | 32.75 | 50.9 | |
| GRPO2026.02 | 32.46 | 51.25 | |
| ClipCov2026.02 | 32.1 | 51.97 | |
| Base ModelModel=DeepSeek-R1-Distill-Qwen-7B2025.10 | 30.85 | 47.31 | |
| BackboneBackbone=Qwen3-4B2026.05 | 30.5 | 40.9 | |
| KLCov2026.02 | 29.5 | 50.18 |