CRITICAL: NVIDIA NIM Chat Completions Timeout — DeepSeek V4.1 Flash

AI & Data Science NVIDIA NIM Access/Accounts
,
1

NVIDIA Hosted NIM API Inference Completely Unresponsive for DeepSeek V4.1 Flash

Dear NVIDIA Developer/NIM Support Team,

I am reporting a critical reliability issue with NVIDIA’s hosted API inference service that is currently blocking development of multiple AI projects.

I am using NVIDIA’s hosted API with:

  • Endpoint: https://integrate.api.nvidia.com/v1
  • Model: deepseek-ai/deepseek-v4.1-flash
  • API: OpenAI-compatible Chat Completions API
  • Environment: GitHub Codespaces
  • Client: Python/OpenAI-compatible SDK and direct curl testing

I have systematically isolated the problem and verified that this is not simply an application implementation error.

Evidence

The NVIDIA API key and network connectivity are working correctly.

A request to:

GET https://integrate.api.nvidia.com/v1/models

returns:

HTTP 200
Response time: approximately 0.035 seconds

However, a direct inference request to the documented Chat Completions endpoint:

POST https://integrate.api.nvidia.com/v1/chat/completions

using:

{
  "model": "deepseek-ai/deepseek-v4.1-flash",
  "messages": [
    {
      "role": "user",
      "content": "Reply with exactly: NVIDIA Brain connected."
    }
  ],
  "max_tokens": 64,
  "temperature": 1,
  "stream": false
}

does not return any response.

The request was allowed to run for 120 seconds and eventually failed with:

curl: (28) Operation timed out after 120001 milliseconds with 0 bytes received

HTTP_STATUS:000

There is:

  • no HTTP response,
  • no response body,
  • no model output,
  • and no useful server-side error returned to the client.

I also encountered an earlier NVIDIA response:

HTTP 500

Missing request extension:
Extension of type
headers::common::authorization::Authorization<...Bearer>
was not found.

Impact

This is now blocking development and testing of my AI projects.

I am building an autonomous AI software-engineering system that depends on NVIDIA’s hosted inference API. The project has a functioning provider layer, correct endpoint configuration, automated tests, and a properly configured NVIDIA API key.

However, because the actual inference endpoint does not return a response, the AI brain cannot be tested at all.

This is not merely an inconvenience.

My development workflow and multiple AI projects are currently on hold because the inference service is not responding.

I have already spent significant development time investigating whether the problem was caused by:

  • API-key configuration
  • environment variables
  • GitHub Codespaces
  • network connectivity
  • incorrect API endpoint
  • incorrect model identifier
  • incorrect SDK
  • incorrect API method
  • request construction
  • application code
  • timeout configuration

The tests have isolated the problem to the hosted inference request.

What I need from NVIDIA

I request that NVIDIA:

  1. Investigate the hosted inference service for deepseek-ai/deepseek-v4.1-flash.

  2. Determine why /v1/chat/completions is accepting the connection but returning zero bytes for 120+ seconds.

  3. Check whether my NVIDIA organization/API key has the required Public API Endpoints entitlement.

  4. Check whether there is an account-level restriction, entitlement problem, regional issue, capacity problem, or backend incident affecting inference.

  5. Confirm whether deepseek-ai/deepseek-v4.1-flash is currently operational through the hosted API.

  6. Provide a concrete workaround if the endpoint is currently degraded.

  7. Provide an

    External Image

    estimated resolution timeline if this is a known service-side issue.

2

I have been having “Unknown Error 504 Gateway Timeout” with that same model every time i try to use it in the past 2 days…

3

I am having timeouts on the same model as well

4

Same here as of 2026-11-06. GET /v1/models returns 200 in about 230 ms, but POST /v1/chat/completions for deepseek-ai/deepseek-v4.1-flash gets no HTTP response within 120 s (tried with several fresh keys). z-ai/glm-5.3-flash hangs the same way for me, while both models answer in the Playground. No 401/403/404, just no reply at all. Is this a known issue on the NIM side?