Model Selection & Routing

Not every task needs the most capable model. A formatting task, a data extraction, or a simple classification can be handled by a smaller, cheaper, faster model. A complex analysis, a nuanced recommendation, or a creative synthesis needs the top-tier model. Model routing directs each task to the appropriate model based on the task's complexity, quality requirements, and cost sensitivity.

The routing strategy: classify each task type by required capability level — low (formatting, extraction, classification), medium (summarization, basic analysis, template-based output), high (complex reasoning, novel synthesis, judgment-dependent recommendations). Route low tasks to the cheapest model that meets quality thresholds. Route medium tasks to a mid-tier model. Route high tasks to the top-tier model. Validate the routing by running each task type through all three model tiers and comparing quality scores. If the cheap model produces quality scores within 5% of the expensive model for a task type, the task routes to the cheap model.