Model deployments for your own models

Your model. Your GPU capacity. One endpoint.

Model deployments on OneInfer Cloud run one model on GPU workers that belong to your deployment, so its traffic does not compete with shared Model API requests. Bring a catalog model, a public or private Hugging Face model, or weights you upload.

Choose the GPU, region and workers in the Console and see the hourly cost before you create it. OneInfer handles provisioning and scaling; you call the endpoint with your API key and its endpoint ID.

Model to endpoint in 5 steps

Step 1 of 5

Choose a model from the catalog, paste a Hugging Face model URL (add your token for a private model), or use weights you uploaded.

What is a model deployment?

It is an isolated model deployment backed by GPU resources configured for your workload. Unlike a shared Model API, its capacity and runtime settings belong to your deployment.

  • Dedicated capacity

    Your requests are sent to the GPU workers assigned to your deployment, avoiding contention with shared API traffic.

  • Predictable configuration

    Control GPU type and count, region, worker limits, concurrency, storage, and scaling behaviour for each endpoint.

  • Private model support

    Deploy supported templates, Hugging Face models, or a custom model source without exposing the model through a public catalog.

Where your requests go

Illustration
  • Without endpoint_idStandard routing to shared Model API capacity.
  • With endpoint_idPinned to the GPU workers of your deployment.
  • As traffic growsScaling adds workers, up to your max workers.

Where your requests go. Without endpoint_id: Shared Model API. Standard routing to shared Model API capacity. With endpoint_id: Your endpoint worker 1. Pinned to the GPU workers of your deployment. As traffic grows: Your endpoint worker 2. Scaling adds workers, up to your max workers.

How many workers run depends on your minimum and maximum workers and the scaling type.

From model selection to a callable API

OneInfer manages the deployment lifecycle while giving your team direct control over the settings that affect performance and cost.

  • Choose the model runtime

    Start from a supported deployment template or configure a compatible model source for your use case.

  • Select GPU and region

    Choose available compute by provider, GPU type, GPU count, and deployment region instead of managing cloud instances yourself.

  • Set workers and concurrency

    Define minimum and maximum workers, scaling mode, and request concurrency to match your traffic pattern.

  • Provision from one console

    Create, inspect, and delete deployments from the OneInfer console without maintaining a separate infrastructure workflow.

  • Call it through OneInfer

    Use your OneInfer API key and endpoint ID to direct inference requests to the exact deployment you created.

  • Build a router

    Attach model deployments or serverless endpoints behind one router for routing, balancing, and failover policies.

Choose the right way to run inference

Read the docs
  • Model APIs for instant access

    Use shared, pay-as-you-go inference when you want to prototype quickly or your traffic is still variable.

    Explore model APIs
  • Model deployments for control

    Use reserved deployment capacity when you need a private model, stable runtime configuration, or predictable production performance.

    Create an endpoint
  • One endpoint for multiple backends

    Group dedicated and serverless endpoints, then route requests according to latency, cost, load, or model policies.

    Explore routing

Create your first model deployment