Pricing
Quoted per workload. Explained before you ask.
No public rate card — deliberately. What we can do better than a rate card is show you exactly what drives your cost, then put the quote in writing before you commit to anything.
Three things decide what you pay.
Model mix
Flagship and open-weight models price very differently upstream. Most production workloads blend both, so a quote reflects your actual mix rather than a blended average that flatters nobody.
Region and isolation
Regions run independent capacity pools with independent economics. Your placement — chosen for latency, data residency or capacity — is a visible line in the quote, not a hidden markup.
Reserved or on-demand
Reserved capacity guarantees throughput and bypasses shared limits; on-demand rides the public pool. Steady base load on reservation with bursts on demand is the usual shape.
Two levers work in your favour at the gateway layer: batch execution and cached input are billed at deep upstream discounts, and cache rates are something a gateway can actively optimise — especially for agent workloads with repetitive context.
Source: Alibaba Cloud Model Studio billing rules · verified 2026-07-26
Start scoped. Grow in writing.
| Shape | What it is | What is included |
|---|---|---|
| Pilot | Fixed scope, one region, your model list | Isolated capacity, request-level reporting, measured with your own traffic |
| Production | Contracted monthly commitment | Committed service terms, per-customer isolation, version pinning, named support |
| Reserved | Production plus dedicated capacity | Guaranteed throughput with graceful fallback to on-demand beyond the reservation |
The commercial fine print, unhidden.
Why is there no rate card on this page?
Because we will not publish numbers we may have to walk back. Regional upstream pricing is still being verified; until that closes, quotes are scoped per workload and put in writing before any commitment.
What does isolated capacity actually mean?
Your throughput is provisioned so another customer’s traffic spike cannot consume it. Shared-pool gateways cannot make that promise. Here it is the default, not an upgrade.
Can we pin a model version?
Yes, and this is where a gateway earns its keep. Upstream, a pinned version carries a far lower request ceiling than a floating alias. We manage that with isolated capacity and region placement so pinning does not quietly cap your throughput.
How do service remedies work?
Committed availability with tiered service credits, written into the contract. Credits are service credits, and the exclusions — including upstream force majeure — are stated plainly rather than buried.
Who signs, and what does procurement need?
A single commercial agreement covers every model in scope. We supply the licence position for each model line, the region and data-handling description for your assessment, and the service terms in one document set.
Get a written quote.
Send your model list, monthly volume and region needs — the quote comes back scoped to exactly that.
hello@steadygateway.com99.9% availability commitment with tiered service credits · Reply within one business day · NDA available on request