Newer NIM models (kimi-k2.6, deepseek-v4-pro) hang indefinitely or 404 — possible missing "Public API Endpoints" permission

AI & Data Science NVIDIA NIM Access/Accounts
,
1

Hi NVIDIA team,

I’m unable to call newer NIM models via https://integrate.api.nvidia.com/v1/chat/completions. Requests either hang indefinitely with zero bytes returned (curl times out after 60s with no response) or return a 404 like:

{“status”:404,“title”:“Not Found”,“detail”:“Function ‘’: Not found for account ‘<account_id>’”}

Models tested:

  • moonshotai/kimi-k2.6 — hangs indefinitely, no response
  • deepseek-ai/deepseek-v4-pro — hangs indefinitely, no response

I’ve confirmed:

  • DNS resolution and base host connectivity are fine (base domain responds instantly with 404 on root path)
  • API key is freshly generated and active (confirmed via account API Keys page)
  • Payload format is correct minimal JSON, tested with max_tokens set low

This matches a pattern I found in other recent forum posts referencing a missing “Public API Endpoints” permission for newer models on personal/free-tier accounts. Could you please enable this permission for my account, or let me know what else might be causing this?

Account email: sdangalla44@gmail.com
Key prefix: nvapi-FjnGMXAr6NCO

Thanks in advance.

2

The symptoms match the “Public API Endpoints” entitlement gating that NVIDIA applies to newer NIM models on personal/free-tier accounts. If that permission is not enabled for your org, the routing layer returns 404 Function not found or simply hangs because the request never reaches a deployed model instance.

What to check:

- Log into build.nvidia.com → Organization settings → look for a “Public API Endpoints” toggle or approval status.

- The base host responding with 404 on / is normal; that only proves DNS works, not that your account is authorized for the model.

- Test an older, generally-available NIM (e.g., an early meta/llama-3.1 variant). If that works and only kimi-k2.6/deepseek-v4-pro fail, it is almost certainly an account-scope issue, not a client bug.

Next step: open a support request through the NIM portal asking to enable Public API Endpoints for your org. Reference the exact model names and the 404 response. Until then, newer frontier models will not be reachable via integrate.api.nvidia.com.

3

No, this is not true.

My account was working fine with Kimi K2.6. It currently works fine with DeepSeek V4.0

This has nothing to do with the account - it has to do with the model.

Please escalate this to someone who knows what’s going on.