Inductive Prolog Rule Synthesis on SLR-Bench Overall 1,000 tasks (full)
86.7Accuracy (%)DF + CC + QO
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| DF + CC + QOCategory=Hybrid (symbolic + LLM fallback)2026.05 | 86.7 | 138.8 | 16.19 | -0.924 | |
| DF + CCCategory=Hybrid (symbolic + LLM fallback)2026.05 | 86.6 | 138.8 | 14.91 | -0.924 | |
| DF + QOCategory=Hybrid (symbolic + LLM fallback)2026.05 | 83.9 | 166.1 | 14.29 | -1.123 | |
| BoK + CCCategory=Hybrid (symbolic + LLM fallback)2026.05 | 80.3 | 132.4 | 14.81 | -0.796 | |
| BoK + CC + QOCategory=Hybrid (symbolic + LLM fallback)2026.05 | 80.3 | 132.4 | 16.09 | -0.795 | |
| DF-32Category=LLM-only methods2026.05 | 79.6 | 224.2 | 17.43 | -0.834 | |
| o3†Category=LLM-only methods2026.05 | 77.8 | 4.3 | 207.24 | — | |
| gpt-5†Category=LLM-only methods2026.05 | 77 | 16.4 | 103.13 | — | |
| BoK + QOCategory=Hybrid (symbolic + LLM fallback)2026.05 | 75.9 | 162.8 | 14.49 | -1.063 | |
| BoKCategory=LLM-only methods2026.05 | 68.7 | 225.3 | 17.88 | -0.611 | |
| CC SolverCategory=Symbolic solvers2026.05 | 68.4 | 0 | 4.01 | -0.756 | |
| QO SolverCategory=Symbolic solvers2026.05 | 60.7 | 0 | 1.28 | -1.087 | |
| Qwen3.6-35B-A3B (OpenHands)Category=LLM-only methods2026.05 | 58.4 | 1,232.6 | 243.41 | -0.354 |