New models in the free API are always overloaded. Why not lower the RPM to 5-10 queries for the most popular models?

AI & Data Science NVIDIA NIM Models
,
1

I understand we shouldn’t expect much from a free API intended for testing, but constant overload makes any use of new models impossible. If providing more computing power isn’t feasible, why not lower the RPM to 5-10? Reducing the limit to these values ​​would have little impact on regular users, but would significantly reduce the potential for abuse of the free API.

This isn’t a complaint; I’m just curious if you plan to address the overload somehow.

2

Kimi 2.5 continues to be overloaded 24/7. It’s frustrating.

3

Same. I can use Kimi K2.5 via the web interface just fine but the API endpoint would not work for some reason.

4

More and more models are becoming overloaded every day. Is Nvidia really not going to do anything about it?

5

It looks like Nvidia has begun efforts to reduce the RPM for high-demand models. Hopefully, this will help make these models more accessible to everyone.