Algorithm Synthesis on Rule-based benchmark 200 contamination-free problems
77CorrectAlgoSkill
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| AlgoSkillBackbone=Haiku2026.06 | 77 | 89 | 100 | |
| Direct LLMBackbone=Haiku2026.06 | 75 | 45 | 95 | |
| CoTBackbone=Haiku2026.06 | 73 | 68 | 99 | |
| Direct LLMBackbone=Llama 8B2026.06 | 69 | 63 | 87 |