Reserving future delivery

Plan a delivery window

Future delivery is an agreement for real inference consumption during a scheduled window. Choose both a start and an end, along with the model, allowance, and throughput requirements.

At the moment you accept, the start must be at least one hour ahead and no more than one calendar year ahead. The server checks this again at acceptance, even if the dates were valid when the quote was created.

Create or review a future request

1. Open Executable offers and select Future delivery to review available offers, or create an RFQ and select future delivery.
2. Choose the exact model and at least 1 BT — one billion input-equivalent tokens. Larger quantities use the same RFQ when the terms need negotiation.
3. Enter the start and end. Check the displayed time zone and UTC offset before confirming.
4. Describe region, retention, token mix, throughput, and any other required constraints.
5. Review the provider’s rate card, capacity, expiry, and delivery terms before acceptance.

A suggestion in a form is not an agreed term. The accepted dates and prices in the agreement control delivery.

Find earlier records

In Executable offers, use Older records and Newer records while the Future delivery tab is selected. Changing between spot and future resets the list to the first page. Requests for quote and Orders & agreements also provide server paging for their records.

The displayed range describes the page you are viewing. Open the agreement itself to confirm its accepted window and status.

A scheduling example

If the current server time is 10:00, the earliest acceptable start is 11:00. A quote for 11:00 that you accept at 10:20 no longer has the required one-hour lead time. Request a revised window instead of relying on the quote’s creation time.

“One year” means one calendar year, not a fixed number of hours. The server applies the boundary, including calendar differences.

Before and during the window

A scheduled agreement is not available for early consumption. At the start time, requests must still meet the agreement’s allowance, route, capability, and throughput rules. At the end, the gateway stops admitting new work under that delivery window.

Monitor the agreement status and remaining allowance in Orders & agreements. Use Usage & reconciliation for recorded consumption and reconciliation states.

Change the model if requirements evolve

You can review an approved replacement quote for eligible unused, unreserved allocation. The amendment preserves unused agreed monetary value and recalculates allowance at the replacement input price. It requires compatible delivery terms and replacement capacity.

Read Switching a future model before accepting a change. A future replacement is an explicit agreement amendment; it is not automatic model fallback.