Skip to main content
Aria Compute hosts device-native model bundles in a public registry. You can browse available models without authentication, but you need an API key or a session token to download bundles. This guide walks you through listing models, choosing a quantization, and fetching the download link.

Before you begin

  • Have an Aria Compute account (any region).
  • Create an API key in the dashboard, or sign in to obtain a session JWT.
  • The base URL for all examples is https://ariacompute.com/api (international). If you are using the China site, replace the host with ariacompute.cn.
1

List available models

Call the public models endpoint to see all bundles and their supported quantizations.
The response includes each model slug, available quant values (int4, int8, int326), and SDK compatibility. Pick the slug and quantization you need.
2

Create or locate an API key

Open the Dashboard API Keys page, then click Create. Copy the full key starting with bfvk-. It is shown only once.
API keys are per site: a key created on ariacompute.com will not work on ariacompute.cn.
3

Request a download link

By default, the download endpoint returns a 302 Found redirect to a short-lived S3 presigned URL. Follow the redirect with -L:
Presigned URLs expire quickly. Do not cache or distribute them: request a fresh link each time you download.
4

Request JSON metadata instead

If you want the URL without immediately following it, send Accept: application/json or add ?format=json.
Use the url field to start your own download. The filename field is the suggested output name.
5

Directory downloads (ZIP stream)

When the bundle is stored as a directory, the endpoint streams a ZIP file with Content-Disposition: attachment. Save it directly:

What you receive

Each bundle contains at minimum:
  • weight.bin — the quantized weights
  • config.json — model configuration
  • Optional tokenizer sidecar files

Errors to expect

Next steps