AI glossary

What is Evals?

Tests for AI behavior: datasets of realistic inputs plus graders that score outputs, run on every change.

Graders can be code-based checks, LLM-as-judge rubrics or human review. Evals catch regressions when prompts or models change.

Related terms