401 Unauthorized on Inference Providers router chat completions (token works for /v1/models)

Hi HF team, I’m trying to use the Inference Providers router for chat completions:

POST https://router.huggingface.co/v1/chat/completions

I consistently get: {“error”:“Invalid username or password.”} (401)

What I’ve already verified:

• GET https://router.huggingface.co/v1/models succeeds (status 200) with the same token.

• Token is fine‑grained with “Make calls to Inference Providers.”

• Inference Providers are enabled in settings.

• I set HF Inference as preferred provider.

• I generated a new token after enabling settings.

Model tried:

• Qwen/Qwen3.5-9B:preferred (also tried other model/provider combos)

The issue persists both in a small Python test and in a Streamlit app.

Could you check if my account has a restriction for chat completions or if there’s another required permission?

Thanks!