What the /v1/models endpoint is, and why it can verify a relay service

What it is

GET /v1/models is a standard endpoint in the OpenAI protocol. It lists the models the current account can use.

curl https://your-endpoint/v1/models -H "Authorization: Bearer your-key"

The response shape:

{
  "object": "list",
  "data": [
    { "id": "gpt-4o", "object": "model", "owned_by": "openai" },
    { "id": "claude-sonnet-4-5", "object": "model", "owned_by": "anthropic" }
  ]
}

Why it is worth checking

1. It proves the endpoint really forwards requests

A site can build a polished landing page claiming support for hundreds of models. That proves nothing -- a landing page is just text.

/v1/models can only return a list if the site actually connects to an upstream. It needs real infrastructure behind it.

2. It exposes the model ids the site actually offers

Naming is not consistent between sites. The same model may be listed as:

claude-sonnet-4-5 / claude-sonnet-4.5 / sonnet-4.5 / claude-4.5-sonnet

The id you get from /v1/models is the one you can actually call, which is more reliable than the name printed in the docs.

3. It makes what is claimed comparable with what is measured

Separating the two is our core method:

A mismatch is normal (marketing is usually more optimistic), but a large gap is worth noting.

The three kinds of response

About 404: some sites deliberately disable this endpoint to stop others scraping their model list. So a 404 does not mean the site is fake, but it does mean the protocol implementation is incomplete and you have lost one way to cross-check it.

How we use it

The models covered figure under measured metrics on a site detail page comes from this endpoint.

If it shows a dash, it means we could not get a list from this endpoint -- not that the site does not support those models, only that we could not verify it.

How you can use it

Before signing up:

curl -s https://your-endpoint/v1/models -H "Authorization: Bearer your-key" | head -c 500

• You get a list, so the protocol implementation is fairly complete
• A 404, so other features may be missing too and you should test more thoroughly
• A 401 or 403 with a correct key is usually normal

After signing up: if the model you want is not in the list, your request will most likely fail. Check the list before you write code.

One caveat

Being in the list does not guarantee a call will succeed.

The list says this model is registered. Whether a call works still depends on whether the upstream channel is available. The list is necessary, not sufficient.

Related reading

• Give a relay service a health check: a 20-minute verification checklist
• OpenAI protocol compatibility: chat/completions vs responses
• How uptime is measured: our methodology and its limits