Create a deployment

Self-serve deployment is still a work in progress and will be available soon. The steps below describe the target workflow but are not yet available for general use. To discuss early access, sign up at console.modular.com/signup.

Inference is served through two kinds of endpoints:

Shared endpoints are serverless and pay-as-you-go, running on multi-tenant infrastructure Modular manages for you. A curated set of models is available this way with no setup — see the supported models page.
Dedicated endpoints are created through a deployment: infrastructure provisioned exclusively for your organization, running a model of your choice from the broader model catalog, with configurable scaling.

This page covers creating a dedicated endpoint via a deployment. You can create deployments via the console or the API.

Deploy a model from the console

The console walks you through model selection, region and scaling options, a pricing estimator, and a final deploy step.

Go to the Models page

Open the Models page in the console and select a model to open its detail page.

Open the deploy dialog

Select Deploy in the top right corner of the model detail page.

Configure and deploy

Choose your region and scaling options, review the pricing estimate, then complete the deploy step to create the endpoint.

Deploy a model with the API

Send a POST request to /v1/deployments to create a deployment programmatically:

POST https://api.modular.com/v1/deployments
Authorization: Bearer $MODULAR_API_KEY
Content-Type: application/json

{
  "name": "my-gemma-deployment",
  "model_id": "google/gemma-4-31B-it",
  "region_group": "us",
  "scaling": {
    "min_replicas": 1,
    "max_replicas": 3
  },
  "configuration": {
    "model_path": "google/gemma-4-31B-it",
    "engine": "max"
  }
}

Field notes:

name: URL-safe slug, max 50 characters. Immutable after creation.
model_id: the model identifier from the console Models page. Immutable after creation.
region_group: one of "us", "eu", or "global".
scaling.min_replicas / scaling.max_replicas: replica count bounds for autoscaling.
configuration.engine: set to "max" to use the MAX inference engine.

A successful request returns HTTP 200 with the deployment object:

{
  "name": "my-gemma-deployment",
  "model_id": "google/gemma-4-31B-it",
  "status": "deploying",
  "endpoint_url": null,
  "region_group": "us",
  "scaling": {
    "min_replicas": 1,
    "max_replicas": 3
  },
  "configuration": {
    "model_path": "google/gemma-4-31B-it",
    "engine": "max"
  },
  "created_at": "2026-04-24T00:00:00Z",
  "created_by": "user@example.com"
}

The endpoint_url is null while the deployment is provisioning. Poll GET /v1/deployments/{name} until status is "running" before sending inference requests. For advanced configuration options (worker sets, scaling behavior, auth), see the API reference.

Inference

Deployments

Administration

Deploy a model from the console

Go to the Models page

Open the deploy dialog

Configure and deploy

Deploy a model with the API

​Deploy a model from the console

​Go to the Models page

​Open the deploy dialog

​Configure and deploy

​Deploy a model with the API

Deploy a model from the console

Go to the Models page

Open the deploy dialog

Configure and deploy

Deploy a model with the API