Documentation/Usage, costs and call diagnostics

Usage, costs and call diagnostics

Review a request using evidence

Review a request using evidence

Inspect model, deployment, status, latency and call ID in call logs. Review usage and the recorded price snapshot in cost details. Do not recompute historical requests with current prices.

Steps

Filter costs by user, key, team or project and check attribution and time range. Cached input is part of input usage. Images, audio and video use their own pricing units; see billing formulas and examples.

Checks and limits

Poll asynchronous tasks with the same identity and credentials. Completed task costs may appear after creation. Preserve task ID, x-litellm-call-id, status and pricing snapshot; a timeout or lost connection does not prove that the provider did no work or charged nothing, and logs are not a complete provider invoice.

Search full documentation: budgets, routing, Docker…