#Usage and Billing

Your final cost depends on the selected model, request type, input and output usage, and any media or task-specific units. Use Models and Pricing for current availability and prices.

#Read usage records

Usage logs can include:

• API key or token used;
• model and endpoint;
• input, cached, reasoning, and output usage when reported;
• media or task units for images, audio, or video;
• request status, latency, and Request ID;
• final charged amount.

Provider usage fields vary, so not every request exposes every category.

#Pre-consumption and settlement

Some requests reserve an estimated amount before the upstream request runs, then settle against actual usage. Temporary balance changes can therefore differ from the final usage log. Failed or interrupted requests can also require settlement after the final provider result is known.

Use the completed usage record—not an in-progress estimate—as the source for a finished request.

#Control spending

• Create one API key per tool.
• Set a key-level quota or expiration when available.
• Start agents with a small model and short test task.
• Monitor unattended tools such as coding agents and OpenClaw automations.
• Limit retries and long-running loops.
• Revoke unused keys.

#Investigate an unexpected charge

Collect the Request ID, time and timezone, API key name, model ID, endpoint, and the relevant usage-log entry. Do not send the full key to support.