Credits and usage

1. 1The request is classified and scored.
2. 2The required credits are reserved atomically, so concurrent requests cannot overspend the balance.
3. 3The request executes.
4. 4Usage is settled at the actual cost and the reservation is closed.

• Model tier — Pro and Deep Strategy use fixed costs; Core is scored per request.
• Context — a large conversation or many connected sources increases the score.
• Research and tools — live web research and tool execution increase the score.
• Output — a large or complex deliverable increases the score.

The cost is computed on the server. The number shown in the interface before you send is an estimate; the settled figure is authoritative.

Each completed Ask response carries a usage summary, and the same figure appears in the request history and the credit ledger. If they disagree, the ledger is authoritative — report the discrepancy.

• Start a new conversation instead of dragging a long thread forward.
• Name the specific source instead of asking OpenFeather to search broadly.
• Reserve Deep Strategy for questions that genuinely need it.
• Avoid re-running requests whose answers you already accepted.

Every completed response carries its own usage summary: what was reserved, what was actually charged, and the tier the request ran under. It is the fastest way to answer "why was this one expensive?".

• Reserved and charged differ when a request failed or finished early.
• The tier explains most of the difference between two similar requests.
• A request charged zero credits performed no billable execution, such as a permission prompt.