Processing limits and queueing
Submit up to 1,000 items; Gizmo schedules them automatically within your Team’s processing limit. You do not need to split a batch to match that limit or choose how many assets run at once. For example, a Team with a 30-asset limit can submit 100 items: up to 30 can process simultaneously, and the rest wait in the queue. All of the Team’s batches and individual generations share that capacity. PollGET /v1/batches/{id} for progress. effective_max_concurrency reports the
current configured processing ceiling, not the number of free slots or a
throughput guarantee. Queued items start as capacity and credits become available.
Cancelling work may continue occupying a slot until its worker has stopped.
Do not resubmit queued items to make them start sooner.
This processing limit is separate from the API-key request rate limit. On HTTP
429, honor Retry-After; an accepted batch waiting in the queue needs only polling.
The optional top-level max_cost_cents must cover the sum of the configured item
budgets. It limits the batch’s authorized customer spend; it does not reserve the
whole amount on submission or change pricing. Actual generation capacity and
pricing policy are configured for the Team, not implied by the 1,000-item limit.
Batch and item responses report reserved_spend_cents (legacy held credits;
new fixed-price jobs do not reserve funds),
settled_spend_cents (customer charges recorded so far),
unresolved_provider_calls and billing_status. Billing remains pending
until work and settlement finish, or reconciliation_required when a charge
needs review. Only complete means the reported customer charge is final.
Generation can be complete while billing still needs reconciliation. Missing
historical accounting returns null amounts and unavailable, never a fabricated
zero. These amounts are customer credit charges, not an all-in Modal/OpenRouter
cost report.
Request usage
GET /v1/usage reports API-key request counts, last-used timestamps and rate
limits. It is not a provider-cost report, credit balance or billing invoice. Its
current implementation reads personal keys; Team usage reporting is pending.
See Errors & troubleshooting for rate-limit headers and retries, and Batch generation for submission examples.