subagent-model-selection

To choose the cheapest capable subagent model, classify the delegated work first to determine required capabilities, then start with the cheapest model tier and escalate only with concrete justification.

Subagent model selection identifies the least powerful model that can reliably finish delegated work, optimizing costs by preferring the narrowest capable agent type and starting at the cheapest model tier.

Apply model selection guardrails before launching any task tool and during parallel delegation to enforce classification prior to model selection and ensure auditable decision records.

The best way to set agent_type and model is to classify the task first, choose the narrowest capable type, and start with the cheapest tier before launching exploration, testing, grading, code review, debugging, or research tasks.

No, you must start with the cheapest model tier and escalate only with concrete justification, ensuring you keep auditable decision records for the subagent model selection.