Autoformalization and Proving on CombiBench (N=100)
96Pass@64Ground truth statement (oracle)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Ground truth statement (oracle)2026.03 | 96 | 68 | 18 | |
| FormalEvolveK=22026.03 | 44 | 27 | 13 | |
| Sample2026.03 | 41 | 23 | 8 | |
| Compile+Semantic RepairSource=Kimina2026.03 | 40 | 23 | 8 |