This is a TLDR of a longer article where I tried multiple free AI models on Cursor and VS Code, through Cline.
The service is called NVIDIA NIM e.g. NVIDIA Inference Microservices.
It lives at build.nvidia.com.
The catalog contains over 100 AI models, many of them free and hosted on NVIDIA's own DGX Cloud infrastructure. The list includes names you already know:
- MiniMax,
- Kimi,
- DeepSeek,
- GLM,
- GPT-OSS, and more.
Get Your Free API Key
Just go to build.nvidia.com and create a free account. You'll need to verify your email and possibly your phone number.
Once you're in, navigate to your account settings and generate an API key. Save it, because you only see it once. It looks like this:
nvapi-xxxxxxxxxxxxxxxx
You get 1,000 free inference credits on signup, with up to 5,000 available on request, and a rate limit of 40 requests per minute. No credit card required.
Use the API key with Cursor
Read about some of the issues with Cursor.
In Cursor, go to Settings → Models, enable the OpenAI API key toggle, paste https://integrate.api.nvidia.com/v1 as the override base URL, and add your nvapi- key.
Use the API key with Cline
Setup takes under two minutes:
- Install Cline from the VS Code or Cursor Marketplace
- Open Cline settings and select OpenAI Compatible as the provider
- Set the base URL to https://integrate.api.nvidia.com/v1
- Paste your
nvapi-key - Enter the model ID from the NVIDIA catalog (e.g. moonshotai/kimi-k2.6). Remember to get an API key for the model.
Top comments (4)
Thanks, this sounds genuinely useful, especially when you're a hobbyist or doing a side project outside of work, and you don't want to spend dozens or hundreds of $$$ just to play around a bit and try things out, or to just learn about AI ...
How does this (Cline plus a 'free' model hosted on Nvidia) compare with something like Claude Code - does it come close in capabilities or quality, or not at all? Well I think you've already (mostly) answered this in your Medium article:
blog.gopenai.com/stop-paying-20-mo...
Brilliant - bookmarked this!
So far, it takes a bit of trial and error to find the right model. As I explained in the full article, I couldn't use some models at all!
They seemed to be too slow, and I am not sure why. Maybe too much demand.
On the other side, gpt-oss seems a good start to try Cline + a free model.
It is not as performant as SOTA models on a paid plan.
Thanks!
I'm sure that a full Claude Code or Cursor subscription delivers more "performance", but obviously also at a higher price (I've seen stories about people burning through their Cursor tokens at an insane rate) ...
If the main goal is to learn about AI and to "play" with it, rather than peek productivity in a corporate/work setting, then a "freemium" setup like this sounds attractive to me ...
P.S. hadn't heard about Cline, it sounds nice!
Yes, I wouldn't call this peak productivity, honestly.
I just went through several models and versions to test them.
Some were unusable, but I got very good results with some others!