Billing

Token models bill recorded input and output usage; request models bill per call. Consult the current pricing table for the applicable unit and rate.

Some models have a minimum charge per call. Short requests can still incur this floor; low token usage does not imply no charge.

Normalized input, output, cache-read and cache-write tokens are all charged at the full current rate. SDK totals may already include cached tokens; do not add that cache portion again. Compare usage fields with your usage logs rather than assuming repeated prompts receive a discount.

Disconnecting after output begins may still incur charges for resources used. Client cancellation does not prove that the provider stopped or that a refund is due.

Reconcile usage, transaction history and request IDs. A displayed balance or missing usage alone does not establish the final charge; include these identifiers when contacting support.

Example: 800 uncached input tokens + 1,000 cache-read tokens + 200 output tokens = 2,000 billable tokens. If the SDK already reports prompt_tokens = 1,800, add only the 200 output tokens, not another 1,000 cached tokens.

Examples receive source review and local checks; this does not establish compatibility with every app version. Check settings, send a small request and confirm usage before production use.