Docs / Limits & pricing

Pricing

You pay per input token; output tokens are free. The current price per million input tokens is on the pricing page.

Each answered request is billed as its content tokens (the state plus every question, counted with the model's tokenizer) plus a small fixed per-request overhead. Rejected requests (4xx) are not charged. usage.input_tokens in every response shows exactly what was billed.

Accounts are prepaid: an organization needs a positive balance to call the API or use the Playground; otherwise requests return 402.

Each request is priced at the rate in force when it ran, so a later price change never alters past spend. Usage and spend are under Usage; credits and history under Settings → Billing.