Code Optimization on HumanEval-Hard Cross-Domain
13.6Calls per TaskTextBFGS
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| TextBFGSKnowledge Base=MBPP KB, Architecture=One-Pass2026.01 | 13.6 | 1,581.7 | 21.6 | |
| TextBFGS-REMOKnowledge Base=MBPP KB, Architecture=One-Pass2026.01 | 13.9 | 1,594.4 | 22.2 | |
| TextBFGSKnowledge Base=None2026.01 | 17 | 1,481.3 | 25.2 | |
| TextGrad-Momentum2026.01 | 29.8 | 1,464.2 | 43.7 | |
| TextGrad2026.01 | 35.8 | 863.9 | 30.9 |