Mathematical Problem Solving on MATH (Training Cost Evaluation)
1.22Training Cost (USD)GDesigner
Evaluation results
| Method | Links | |
|---|---|---|
| GDesignerModel selection=Defaults to GPT-5-Mini for all agents2026.02 | 1.22 | |
| MASRouter2026.02 | 1.97 | |
| HieraMASLLM Selection=Cost-aware rewards, Optimization=Two-stage algorithm with graph classifier training2026.02 | 2.03 | |
| AFlowOptimization=Monte Carlo Tree Search, Base model=Claude 3.5-Sonnet2026.02 | 18.62 |