Dedicated Endpoints

Private GPU deployments of open-source models, called through the same chat/completions interface.

Dedicated Endpoints let you spin up a private GPU deployment of an open-source model and route inference to it through the standard /chat/completions interface. Deployments are isolated to your account, have no request-level rate limits imposed by the shared router, and support autoscaling between a minimum and maximum replica count.

Replace {organization} in the base URL with your workspace slug. For example, if your organization is Uber, requests go to https://uber.blackbox.ai.

Deployments are billed per GPU-minute while running. Use min_replicas: 0 on the scaling object to let idle deployments scale to zero.

Two ways to address a deployment

Once a deployment is created you receive a client-facing model id of the form dedicated/<sanitized-slug>-<12char-suffix>. You can pass either the full id or the bare slug (without the suffix) as the model field in a chat completion request.

1. Explicit full id

Routes to that specific deployment. Recommended for production so requests never fall back to a different replica pool.

curl -X POST https://{organization}.blackbox.ai/chat/completions \
  -H "Authorization: Bearer $BLACKBOX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "dedicated/openai/gpt-oss-20b-Ab3xY9K1mN2p",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

2. Bare slug (auto-pick)

Drop the trailing suffix and the router auto-picks one of your deployments for that model. Useful during development when you have a single deployment per model and don't want to hard-code the suffix.

curl -X POST https://{organization}.blackbox.ai/chat/completions \
  -H "Authorization: Bearer $BLACKBOX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "dedicated/openai/gpt-oss-20b",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

Auto-pick only considers deployments owned by the caller's API key. If you have zero deployments for the model, the request returns 404 deployment_not_found.

Endpoints

Endpoint	Description
`GET /dedicated/templates`	List all models you can deploy, along with valid GPU/region combinations.
`POST /dedicated/create-deployment`	Create a new deployment from a template.
`GET /dedicated/list`	List your deployments (and the templates they were built from).
`POST /dedicated/edit`	Update scaling, description, or enabled state of a deployment.
`POST /dedicated/delete`	Permanently delete a deployment.

Quick start

List available templates

Call GET /dedicated/templates to discover which models you can deploy and which gpu_type, gpu_count, and region values each template accepts.

Create a deployment

Call POST /dedicated/create-deployment with a model_name from the template list, a valid gpu_type / gpu_count / region, and optional scaling. The response contains the client-facing model id.

Send chat completions

Point your existing chat-completions client at https://{organization}.blackbox.ai/chat/completions and pass the deployment model id (or bare slug) in the request body. All standard chat-completions parameters are supported.