Before you begin, send representative requests through the task. Configure model prices when Studio can’t identify them automatically so cost comparisons have sufficient pricing coverage.
1
Select a time range
Open the task’s Analytics page and choose a range from one hour through three months. Use a range that represents the traffic you expect the candidate to serve.
2
Check traffic health
Review request volume, routed model share, task versions, token volume, and provider errors. Resolve unexpected errors or routing changes before comparing costs.
3
Check pricing coverage
Confirm that the models you want to compare have similar pricing coverage, so the cost comparison in the next step reflects both models accurately. Add a manual price in provider or fine-tune settings when Studio can’t find a price.
4
Compare model costs
In the cost calculator, select a comparison model. Review the estimated cost and savings if all traffic in the selected range had used that model.Treat the result as an estimate. Use evaluations and application metrics to confirm that a less expensive model preserves the quality your task requires.
5
Monitor a rollout
After changing routing, confirm that the candidate receives the intended share, errors remain stable, and estimated cost moves in the expected direction.
Analytics metrics and pricing
Learn how Studio calculates traffic, token, error, and cost views.