NVIDIA team,
I’m a NVIDIA Developer Program member using the free serverless NIM inference
endpoint. I generated a fresh API key (nvapi- prefix) from
build.nvidia.com/settings/api-keys and also tried creating one directly from
the model page. The key is valid (calling api.nvcf.nvidia.com returns 404, not
401/403), but every call to the Chat Completions endpoint returns:
POST https://integrate.api.nvidia.com/v1/chat/completions
→ HTTP 403 {“status”:403,“title”:“Forbidden”,“detail”:“Authorization failed”}
This happens across multiple models (including meta/llama-3.3-70b-instruct),
from both curl and an OpenAI-compatible client, so it’s not model-specific or
client-specific.
The key was tested within a minute of being minted and 403s instantly, which
rules out activation delay or quota exhaustion. Based on the many similar
reports in this forum, it appears my personal organization is missing the
“Public API Endpoints” permission.
Could you please enable the “Public API Endpoints” permission for my account?
Thank you.