Skip to main content
The serverless catalog is a curated set of popular open models. You pick one in the dashboard. There are no workers to size, no weights to fetch, and no containers to build. Fireworks AI hosts these shared models. That is the difference from launching your own model. A catalog model is a shared deployment. It is not dedicated to your account. Other clients call the same model. Your own launch is capacity you configure: workers, scaling, and the weights or image you choose. Use the catalog to prototype quickly. Launch your own model when you need dedicated capacity, custom weights, or other pipelines.

Call the API

Authenticate and send your first request.

Migrate from OpenAI

Change the base URL and API key; keep your existing client.

How deployment differs

Exact rates are shown under Live pricing on the Usage page. Do not hard-code dollar amounts from this page.

Why use the catalog

  • No infrastructure to set up. Select a model and send a request. You do not size workers, scale them, or build an image.
  • A short, ready list. The models and pipelines are the ones in the catalog.
  • Usage-based. Catalog usage is paid from your credits and shows up in Usage.

Available models

The catalog changes as models are added or retired. The list you see in the dashboard when you add a model is authoritative.

Add a catalog model and call it

Adding a catalog model gives you a model id in your account. The deployment behind that id is shared, not dedicated to you, and you do not configure workers. Use the model id from the dashboard in your requests.
1

Open the launch wizard

In the dashboard, start a new model, as described in Launch a model.
2

Choose the serverless catalog

Select the catalog as the source, then pick the model you want from the list.
3

Confirm and copy the model id

Once the model is added, open its page in the dashboard and copy the model id. See Manage models.
4

Create an API key

Create a key on the API keys page and make sure your account has credits. See API keys and Billing.
5

Send a request

Use the model id from the dashboard as the model.
To stream the response, see Streaming.
Try the model in the Playground first, then paste the request into your app.

Data handling

Catalog models are shared deployments, not dedicated to one client. Fireworks AI hosts them. See Shared catalog in the Data Policy. Catalog requests are processed in the United States by Fireworks, only for the time needed to generate and return the response. Per Fireworks’ terms, inference content is not used to train its models and is not kept after the output is delivered, except where law requires it. SynapsAI does not use your content to train models or sell it. If you enable Store requests, matching requests appear on the Data tab for your account only. See Data and the subprocessor list.
If you need inference to run only on infrastructure you control, launch your own model or see Deployment modes.

Limits

Catalog requests follow the default rate limits for inference. Need more? Contact support or ask about quotas.

Troubleshooting

  • Model not found. Check that you are using the model id from the dashboard, not the catalog model name. Catalog models can be retired.
  • Insufficient credits. Add credits or turn on automatic top-up in Billing.
  • Other errors. See Troubleshooting and the status page.

Next steps

Call the API · Examples · Launch a model · Optimizing costs