Loading
LoadingCheaper Inference / Enterprise / Evaluation
A small, representative comparison is easier to review than an all-at-once migration—and gives finance and engineering the same source of truth.
Start with a model or traffic slice whose current cost and expected behavior are already understood.
Keep the familiar request shape, then point the compatible client at Cheaper Inference with a workspace API key.
Compare settled cost, savings attribution, token use, request status, and reliability before expanding traffic.
Start with the public catalog and documented API. The same workspace then carries the keys, usage records, and billing evidence needed to review the result.
The application record, optional caches, and the provider serving a request have different data boundaries. Here is where each one starts and ends.
Core metering records retain model, endpoint, token counts, charges, status, and sanitized failure details. They do not include prompt or response bodies; optional caching is a separate data boundary described below.
Account response caching can store request and response pairs within that account and can be switched off. Provider prompt caching may also create provider-side cached state.
Inputs are sent to the provider selected to serve each request. That provider may process or retain content under the contractual settings and retention practices applicable to its service.
We do not make a universal zero-data-retention claim. Confirm exact route and provider coverage with security before sending production data.