March 31, 2026
NVIDIA Build Is Surprisingly Good: Use It with OpenCode to Write Code for Free
Today we are not talking about programming.
By Donma Hsu 當麻許
6 min read
Well… not exactly.
Let's talk about something useful instead.
If you are not using ChatGPT Codex because it costs money, but you still want to experience the joy of having AI help you write code, then this is something worth trying.
Here is a combo I think is genuinely worth recommending. Not many people seem to be talking about it, but after playing with it for a bit, I think it is pretty solid:
1. Sign up for an NVIDIA developer account
First, go to https://build.nvidia.com and create a developer account.
It is not hard.
When you request the API key, you will need a phone number to receive a verification code. In my case, I used a Taiwan mobile number.
After that, just click Get API Key and copy the key.
That is step one done.
2. Download OpenCode
Next, download OpenCode from here:
Just pick the version for your operating system.
I tested it on Windows.
After using it for a few days, one thing I noticed is that it gets updated really fast. It feels like there is a new version every few days.
Which honestly makes sense. AI tools are evolving at a ridiculous speed lately, so seeing a project get updated this often is actually a good sign.
At least it feels like somebody is seriously maintaining it.
3. Add NVIDIA as the provider in OpenCode
Then go to the settings in the bottom-left corner.
From there:
choose Providers
click View more providers
select NVIDIA
paste in the API key you got in step one
That is basically it.
Nothing too fancy here.
4. This is the part a lot of people miss
And honestly, this is the fun part.
A lot of people do not even realize how many models are available.
You can click Manage Models and turn on the ones you want to use.
Once you enable them, you can switch between a bunch of different models depending on what you need.
This part is easy to overlook, but it is worth checking.
I was surprised that some of my friends had no idea this was even there.
5. Bind it to your project folder and start using it
After that, it is just normal usage.
Bind OpenCode to the project folder you want, and now AI can start helping you work on your code.
That is really the whole idea.
Just a pretty straightforward setup that gets out of your way.
Which models are actually worth using? There are a lot of options, but when you are using OpenCode for real coding work, you probably will not touch all of them. The ones that feel more useful from an engineering point of view are usually Qwen Coder, DeepSeek V3, Codestral or Devstral, Kimi K2.5, GLM 4.7, Llama 3 or 4 Instruct, Nemotron, Granite Code, and Phi..
That said, they are all there if you want to try them, and honestly, 40 RPM is more than enough for everyday coding. If you are mainly asking questions, fixing bugs, generating snippets, refactoring code, or letting AI help you explore a project, it works just fine. It also helps a bit with that lovely modern problem called token anxiety — that feeling where every prompt sounds like it is charging rent.
That is it for now.
Happy Coding — or maybe, Happy Commanding.
Originally published at https://donmatw.blogspot.com.