Skip to main content
A launch creates a private deployment for almost any model we list under Supported tasks — chat and completions, but also embeddings, images, audio, video, classification, QA, and the rest of that table. The dashboard route is Launch a model. Buttons often say Deploy. This page covers managed inference. For your own container, see Custom image.

​
Before you start

Public Hub repos launch without a token. Gated or private repos need a Hugging Face token (account-wide or model-only in the wizard). Safetensors weights are supported. Custom Python is supported only the way Transformers, Diffusers, or vLLM load remote code. Prep a repo in Hugging Face setup.

​
Wizard steps (managed)

  1. Type — Managed model inference.
  2. Model — Hugging Face repo + revision, or pick an artifact.
  3. Configure — workers, timeout, scaling, precision, quantization, pricing, store requests.
  4. Review — confirm the live estimate, then launch.
Empty Models lists also offer Deploy an example presets.

​
Settings

The wizard shows estimated storage and compute before you confirm. After launch, status moves Building → Sleeping or Ready. See Manage models.

​
Next steps

Playground

Try the deployment in the dashboard.

Call the API

SDK, cURL, and OpenAI-compatible clients.

Integrations

Drop the deployment into LangChain, LlamaIndex, or FastAPI.

Autoscaling

Tune workers after launch.