How do I use an OpenAI-compatible API for open models?

CloudGPU provides an OpenAI-compatible API at https://cloudgpu.app/v1. Configure the base URL and API key, then select an open-model ID supported by the endpoint.

The model list from that endpoint is the set of IDs it accepts. It is not a statement that every returned model is an open model. Choose an ID you know is an open model and that appears in the list.

Base URL, key, and model discovery

Set the base URL to https://cloudgpu.app/v1. Create a key in the API console and send it as a Bearer token from the environment variable CLOUDGPU_API_KEY. Do not commit a real key. In the samples below, your-open-model-id is a placeholder, not a model ID, and the commands were not run for this page.

List models:

curl -sS "https://cloudgpu.app/v1/models" \
  -H "Authorization: Bearer ${CLOUDGPU_API_KEY}"

Chat completions request shape:

curl -sS "https://cloudgpu.app/v1/chat/completions" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer ${CLOUDGPU_API_KEY}" \
  -d '{
    "model": "your-open-model-id",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

Request and response fields are on the API docs. Current per-token rates stay on the API page. This page does not copy them. The same prepaid balance pays for this API and for GPU rental. Payment methods are on payments.

What compatibility does not mean

A request in this shape is compatible at the request-format level. That does not guarantee feature support, model quality, tool calling, or equivalence to another service.

This is not vLLM on a rented GPU

The calls above go to CloudGPU’s hosted gateway at https://cloudgpu.app/v1. They do not start a GPU. To install vLLM on a machine you rent, follow how to serve an OpenAI-compatible API with vLLM on a cloud GPU and pick a card on Deploy.

To enter the same hosted base URL in Cursor, see how to use CloudGPU as a custom base URL in Cursor.

Frequently asked questions

What base URL do I use?

https://cloudgpu.app/v1

How do I choose a model ID?

Call GET https://cloudgpu.app/v1/models and select an open-model ID that the endpoint returns. The catalog does not mean every listed model is an open model. your-open-model-id in the sample is a placeholder, not an ID to send.

Does a compatible request guarantee every feature?

No. Request-format compatibility does not guarantee feature support, model quality, tool calling, or equivalence to another service.

Sources (read 2026-11-11)