Dedicated Endpoints
Your own replicas of a model, on GPUs no one else uses, behind the same OpenAI-compatible API.
Model replicas on GPUs reserved for your workload.
Your model. Your GPUs.
Run a catalogue model or your own weights on single-tenant GPUs in a named UK or EU region.
Private networking connects directly to your existing estate. Agree the model, region and capacity around your workload.
On demand or reserved
Run on demand, billed per minute while deployed, or reserve capacity for a committed term.
- On-demand endpoints scale to zero, with free start-up time
- Reserved services carry availability terms agreed in your Order
- Choose whether excess traffic overflows to the Token API or is capped
Built for production
For teams scaling past shared-fleet noise, running custom fine-tunes or serving products where users feel latency.
Contractual no-training terms and a named jurisdiction help your procurement and data-protection teams assess the deployment.
What comes next
Per-endpoint evidence packs and pinned-version change notice are in build. Provisioned-throughput packaging is planned for a later phase.
Ask us for the current status and dates for your deployment.