Requesting and comparing quotes
Turn your requirements into a provider agreement.
What an RFQ does
A request for quote, or RFQ, tells suppliers the inference service you want to purchase. It does not reserve capacity or create a purchase by itself. A provider responds with an expiring quote and rate card for you to review.
Use an RFQ when you need specific delivery terms, want to compare providers, or cannot find a suitable firm offer.
Find an existing request
The Requests for quote list loads records from the server in pages. Use Older records for the next page and Newer records to return. Open the request reference to inspect its requirements and provider quotes.
Orders & agreements uses the same controls for accepted records. A page range is not a lifetime total, and permissions are checked again when each page is loaded.
Create a request
Open Requests for quote, then choose New RFQ. You can also start from a model’s Request a quote action, which carries the model into the form.
Complete the following:
| Field | What to enter |
|---|---|
| Model | Type a name or ID and select a matching catalogue model; arrow keys and Enter also select a result |
| Quantity | At least 1 BT, meaning one billion input-equivalent tokens. A firm offer is the instant path; an RFQ is for negotiated terms |
| Delivery type | Spot for immediate delivery or future for a scheduled start |
| Start and end | Explicit delivery window |
| Required tokens per minute | Optional buyer requirement. Enter your workload's minimum rate; observed published rates for the selected model appear as suggestions, not promises of delivery |
| Quantity flexibility | Exact requested quantity by default. Choose 10%, 20%, or 30% only if you would accept a different quoted quantity. A new purchase still cannot be below 1 BT |
| Preferred region | Choose No preference or a region recorded on a qualifying published offer for the selected model |
| Data retention | Choose No preference or a retention policy recorded on a qualifying published offer for the selected model |
| Preferred provider | Choose No preference or a provider on a published offer for this model. A preference restricts matching quotes; it does not reserve that supplier |
| Execution mode | Choose No preference, or a mode shown in the published offers. A batch requirement needs a provider that confirms batch support and both batch prices |
| Quote deadline | Optional local date and time after which providers may no longer respond |
| Workload and delivery requirements | Write your own Markdown notes about input and output mix, throughput, and constraints. Use Preview to read the formatted text and Edit source to change it |
No preference leaves a choice unspecified. Changing the model clears provider, mode, region, and retention choices. An empty list means the current published-offer snapshot has no additional options; an RFQ can still ask providers for terms. The page reports when supply evidence could not be loaded. It may show only a sample of offers, while the choice lists cover every currently eligible published offer. Your selections are requirements, not confirmations that a provider can fulfill the combination. Assistant advice appears separately and is never submitted as your notes unless you enter it yourself.
While a request is open, its status is green in the request list and on the overview. Open the request from either place. The buyer can edit an open request until a provider quote is recorded. Each save is kept as an edit, with the editor and the previous and new terms, including changes to Markdown notes. Providers and brokers see that history on the request before they quote. Editing an existing spot request keeps its recorded delivery window unless you change the delivery type.
Choose Review settings with AI for advice using your current settings, catalogue records and published supply. The recorded decision tree shows evaluated excerpts and scores. Recommendations never apply settings or submit an RFQ automatically. If you change a field, the previous review is marked stale. Scores are bounded assessments, not guarantees of correctness or reserved capacity.
A broker must also select an authorized buyer organization. The request belongs to that buyer’s procurement workflow.
Compare returned quotes
Open the RFQ to view its quotes. Compare input and output prices separately. Review any cache rates, additional supported units, tiers, delivery conditions, and the supplier’s terms.
Check the quote’s revision and expiry. A superseded or expired quote should not be treated as a current offer. The latest accepted terms become the agreement record.
Providers responding to an RFQ select a capacity pool that covers the requested model and full delivery window, then enter the allowance, rate card, expiry and terms they are offering. A partial quote may be smaller than 1 BT if the accepted combined agreement still meets the minimum. Submission rechecks current capacity and the provider's authority.
Combine partial provider quotes
A provider can quote all of your requested allowance or offer a smaller allocation. Read Quoted allowance on each quote. A partial allocation may be below 1 BT; the complete new agreement must still meet the 1 BT minimum.
For a quote that covers the full RFQ, use Review & accept quote. To combine allocations:
- Find Combine provider allocations on the RFQ detail page.
- Select each provider quote you intend to include. Each row names the quote, allowance, provider, and expiry.
- Check Selected / Required. By default, the selected allowances must add up to the full requested quantity exactly; when the RFQ specifies a Quantity flexibility tolerance (10%, 20%, or 30%), the combined total may fall within that tolerance band as long as the complete new agreement is at least 1 BT. Select between one and twenty quotes, without duplicate entries.
- Review the rate card and terms shown for every selected quote. A lower input price on one allocation does not change another allocation’s prices.
- Select Accept selected provider allocations only after the complete selection matches your requirement.
The server checks current quote revisions, expiry, buyer authority, commercial approval, policy, and each supplier’s capacity. If any allocation is unavailable or invalid, the whole acceptance fails without creating a partial agreement. An old selected browser view does not guarantee current capacity.
For example, a 10 BT RFQ could be covered by a 6 BT quote and a 4 BT quote. Selecting only 6 BT, or selecting 6 BT plus 5 BT, does not match the request. These are illustrative quantities, not live offers.
Accept deliberately
Before accepting, verify the selected provider capacity, model, allowance, and delivery window. Your organization needs commercial approval, and you need authority to accept. A broker additionally needs an active mandate with acceptance rights.
On acceptance, the service checks the current time, quote expiry, and available capacity and creates the agreement atomically. Repeating an acceptance request with its same operation identity returns the original result rather than creating a duplicate purchase.
Leaving the page or closing a dialog after submission does not undo an acceptance already sent. The request may still finish, but its response will not take you away from the page you chose. Check Orders & agreements for the recorded result. If the dialog reports an uncertain failure, reopening the same acceptance and retrying keeps its operation identity.
When a request needs changing
If a requirement or quote is unsuitable, create a clearly revised request or ask the provider for revised terms before acceptance. Do not accept a quote based on an assumption that a price, model, region, or delivery date can be changed afterward.
Inspect the resulting allocations
After successful acceptance, open Orders & agreements and select the agreement. For a multi-provider agreement, Provider allocation rate cards shows each supplier’s model, input-equivalent allowance, remaining and reserved value, delivery dates, and exact accepted prices.
Use these allocation cards when assessing provider-specific consumption. The overview rate card is not a universal price for all providers. The complete record is also available under Provider allocations and capacity reservations.
An accepted future model replacement follows a separate amendment workflow with its own capacity and value checks.