alt.hn

4/2/2026 at 5:13:59 PM

Show HN: I tested 15 free AI models at building real software on a $25/year VPS

https://georgelarson.me/writing/2026-04-03-25-dollar-ai-lab/

by j0rg3

4/3/2026 at 1:40:38 AM

> Instead of fixing the rate limiter that was blocking its own tests, it patched the environment detection. That's enterprise development in a nutshell.

That part made me laugh and reflects my experience when I was working on "enterprise development" teams.

I'm curious about having this run on a VPS as opposed to a local VM. What does that provide you? I understand having the VM completely disconnected from your local network etc but is there anything besides that? I ask because mentioning the cost here seemed like one of the important points but that cost wasn't necessary for the experiment itself.

by scorpioxy

4/3/2026 at 12:15:54 PM

I’m a little confused. So you got a VPS… was that just to host the result? Surely you weren’t running the models on it?

What were the models running on? It doesn’t matter a ton but would help paint the whole picture better.

by bentt

4/4/2026 at 1:43:27 PM

Also no mention of quants and not even sizes in many cases. I wish someone did this kind of comparison well.

by a96

4/5/2026 at 1:05:21 PM

Opencode doesn't publish what quantization they offer.

by polski-g

4/3/2026 at 12:33:08 AM

Curious how the top performers compare to SOTA paid models. Also interested in understanding performance differences between running model on VPS hardware compared to a laptop. Thanks for sharing

by walkersutton

4/3/2026 at 1:34:00 AM

I don't believe the models were running on the VPS itself. According to my understanding of the article, they used OpenRouter and OpenCode's service as service providers. The agent was the thing running on the VPS.

by scorpioxy

4/4/2026 at 2:54:17 PM

What VPS is $25/year? DigitalOcean seems to be about 3x that price for the low end droplets…

by BobbyTables2