Pay for what you ask.
One price per input token, billed from prepaid credits. No plans or seats: every request shows exactly what it was billed.
What is billed
Its input tokens. Output tokens are free.
The state plus every question, counted with the model's tokenizer, plus a fixed 250 tokens per request. Every response returns usage.input_tokens: exactly what was billed.
327 tokens · $0.0000137
The two-question TSLA request on the home page was billed 327 input tokens, overhead included: about 72,800 requests like it for $1.
Rejected requests
A request that is rejected with a 4xx (invalid JSON or questions, a bad key, a rate limit, no balance) is not charged.
The API, the Playground and workflows
Playground runs are billed like API calls, and every symbol in a workflow run is metered like a request.
Prepaid credits
An organization needs a positive balance to call the API or use the Playground; without one, requests return 402. Usage, spend and remaining credits are in the console, per organization and shared by all of its keys.