Home → Help

A custom endpoint's models do not appear in the client's list

What you are seeing

• The model dropdown is empty after switching to a custom endpoint
• Only the provider's built-in model names are offered
• A model you know exists is not selectable
• Typing the model name manually works, picking it from the list does not

Why it happens

Clients that hard-code model names were built against one provider. Pointing them elsewhere gives you the right endpoint with the wrong menu, and the names they send may not exist upstream at all.

Clients that do fetch the list usually call GET /v1/models. If your base URL is missing the version prefix, that call 404s silently and the dropdown stays empty with no error shown.

Some clients cache the list aggressively. A model added upstream after you configured the client will not appear until the cache is cleared or the app is restarted.

Confirm it is this

Ask the endpoint directly — this is what a well-behaved client sees:

curl -s 'YOUR_BASE_URL/models' -H 'Authorization: Bearer YOUR_KEY' \
  | head -c 400

A JSON list means the endpoint is fine and the client is the problem. A 404 means the base URL is wrong; HTML means it is pointing at a website.

How to fix it

1. Type the model name manually where the client allows itMost clients accept a free-text model field somewhere even when the dropdown is fixed. That bypasses the list entirely.
2. Use model mapping when names cannot be changedIf the client insists on its own identifiers, map them onto names the endpoint serves. Some client-side tools expose this directly.
3. Restart after changing the endpointConfiguration and model lists are usually read at launch. Closing the window is not always enough — quit the application fully.

Related

Last checked 2026-10-01. Written from problems diagnosed on a live OpenAI-compatible gateway, not collected from other sites.