Code Reasoning on HumanEval+
97Pass@16ResRL
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ResRLBackbone=Qwen3-4B2026.05 | 97 | — | |
| NSRBackbone=Qwen3-4B2026.05 | 96.9 | — | |
| CaSPModel=DeepSeek-R1-Distill-Qwen-7B2025.10 | 96.32 | 83.9 | |
| LAD2026.02 | 95.71 | 82.29 | |
| GRPOBackbone=Qwen3-4B2026.05 | 95.7 | — | |
| FlowRLBackbone=Qwen3-4B2026.05 | 95.7 | — | |
| DAPOBackbone=Qwen3-4B2026.05 | 95.7 | — | |
| EntAdv2026.02 | 95.09 | 81.05 | |
| KLCov2026.02 | 95.09 | 80.75 | |
| FlowRL2026.02 | 95.09 | 81.29 | |
| ClipCov2026.02 | 94.49 | 81.48 | |
| Base ModelModel=DeepSeek-R1-Distill-Qwen-7B2025.10 | 94.48 | 81.02 | |
| GRPOModel=DeepSeek-R1-Distill-Qwen-7B2025.10 | 94.48 | 80.25 | |
| GRPO2026.02 | 93.25 | 80.9 | |
| BackboneBackbone=Qwen3-4B2026.05 | 89 | — |