Skip to main content
All terms
Evaluation

A/B Test

Comparing two versions by routing traffic to each and measuring which yields better outcomes.

Definition

An A/B test compares two versions of a system — such as two prompts, models, or interfaces — by routing real users or tasks to each and measuring which produces better outcomes. Because both versions run under the same conditions, differences in metrics can be attributed to the change. It is a common way to validate improvements before a full rollout.