Issue with 429 errors on NVIDIA NIM despite low usage

AI & Data Science NVIDIA NIM Models
1

Dear NVIDIA,

First of all, I would like to thank you for offering the community a free tier for NVIDIA NIM.

However, I’m currently running into some issues and would like to know whether these are intentional API limitations or unexpected behavior.

I initially sent two requests in quick succession to the GLM 5.2 model, and both worked fine the model responded very quickly. But when I tried a third request, I received a 429 error (Too Many Requests). I tried spacing out my requests by 2, 4, 8, 16 seconds, and even longer intervals, but I kept getting the same error. Eventually, I decided to wait about five hours and try again with just a single request and I still got a 429.

I then switched to the MiniMax-m3 model. I was able to send four requests successfully, but the fifth one again returned a 429 error. Now I’m unable to use either model.

I understand that certain models may have flexible limits throughout the day depending on server load. However, I’ve heard that the basic limit is around 40 requests per minute, so I’m a bit disappointed to find myself limited to just 2-3 requests over nearly half a day and I have no idea when the limit resets.

I’m wondering whether this is a system ban due to something I’m doing wrong, or if this is expected behavior. I would also like to ask if there is any place where I can check the approximate request rate limits per minute for each model. Please point me to any documentation or resources I might have missed. Otherwise, I’m afraid NIM may not be usable for me in its current state.

If nothing can be done, could you suggest an alternative model that performs best under these circumstances?

Thank you in advance for your time and response.

I’m new to NVIDIA NIM and really hope we can have some fun together! :)

2

Nvidia seems to have changed the limits. Unfortunately I don’t think there’s an official and reliable way to check the limits. Considering it apparently varies by user, I would say there’s a token limit per day on certain models now. Some models have degraded as well.