Statement generation on CombiBench N = 100
1CH@100Sample
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| SampleBaselines=Kimina, Archive=None, Model=Kimina-7B2026.03 | 1 | 0.44 | 0.816 | 63.7 | |
| Hybrid controlArchive=None, Generation model=Kimina-7B, Repair model=Qwen3-30B-A3B2026.03 | 1 | 0.53 | 0.79 | 58.8 | |
| FormalEvolve (K=2)K=2, Repair model=Qwen32026.03 | 1 | 0.58 | 0.759 | 53.1 | |
| FormalEvolve (K=1)K=1, Repair model=Qwen32026.03 | 1 | 0.55 | 0.726 | 44.2 | |
| FormalEvolve (K=2; w/o EvolAST)K=2, EvolAST=False, Repair model=Qwen32026.03 | 1 | 0.5 | 0.776 | 57.7 | |
| FormalEvolve (K=2; w/o patch repair)K=2, Patch repair=False, Repair model=Qwen32026.03 | 1 | 0.47 | 0.772 | 49.4 | |
| Compile RepairBaselines=Kimina, Archive=None, Model=Kimina-7B2026.03 | 0.99 | 0.4 | 0.825 | 64.1 | |
| Compile+Semantic RepairBaselines=Kimina, Archive=None, Model=Kimina-7B2026.03 | 0.99 | 0.46 | 0.813 | 60.9 | |
| Qwen3 + Compile+Semantic RepairBaselines=Qwen3, Archive=None, Model=Qwen3-30B-A3B2026.03 | 0.94 | 0.39 | 0.827 | 65.5 | |
| Qwen3 + Compile RepairBaselines=Qwen3, Archive=None, Model=Qwen3-30B-A3B2026.03 | 0.9 | 0.26 | 0.74 | 38.5 | |
| Qwen3 SampleBaselines=Qwen3, Archive=None, Model=Qwen3-30B-A3B2026.03 | 0.3 | 0.2 | 0.939 | 95.2 |