Pricing and limits
| Use | Price |
|---|---|
| Decisions | $0.06 per million input tokens; output is free |
| Training | $0.20 per million training tokens, with a $2 minimum per run |
Billing is prepaid. Top up with $5, $20, $50 or $100; all API keys on an account share its balance. The first paid top-up includes one training run: it covers the $2 minimum without using purchased credit, and tokens above that are charged. There are no subscriptions or automatic recharges.
What counts as an input token
A decision request is charged for its logical input: the shared state once, plus each question once. Usage records show the billed input tokens for every request.
A training job reserves the $2 minimum when it leaves uploading. Validation then measures the job's workload, and the reservation rises to the full price before training starts. If your credit cannot cover it, the job fails with the price in its error and nothing is charged. Image training jobs stay at the minimum. Infrastructure or validation failures release the reservation.
When credit runs out
New requests return HTTP 402 with code insufficient_credit. Add credit and retry. Requests already admitted are not interrupted.
Limits
z3-mini limits lists the token, outcome and image limits. Oversized inputs are rejected, not truncated. Throughput limits return HTTP 429 with a Retry-After header.