Drift is a gradual change in behavior, quality, or cost over time even when your code has not changed. It comes from provider model updates, shifting input distribution, or prompt changes. Detect it with evaluators on sampled production traffic and alert when scores slip.

Why it matters

A model can be "up" while quietly getting worse or more expensive. Drift is the slow failure that uptime monitoring misses entirely.

How it works

Drift comes from provider model updates, changing input distribution, or prompt edits. Detect it by running evaluators on sampled production traffic and watching scores over time.

Example

Judge scores on live traffic slip from 4.6 to 4.1 over three weeks after a provider model update, before any user complaint.

← Back to the full glossary

Put the platform behind the terms

Route, evaluate, and monitor every AI request from one OpenAI-compatible platform.

Start Free → Explore the Features