Software developer and AI tinkerer in the UK. I run a homelab that is slightly too ambitious, price-check every AI model once a month and write down what actually worked.
Every month I take the same basket of real tasks, run it through the models I can reach, and write down the cost per job, not per token. This month the gap between the frontier and the free tier narrowed for coding, and widened for anything that needs a long memory.
Read the October report →Two NAS boxes, a couple of small servers and a laptop doing more than any laptop should.
It hosts my own Git server, a CI box, a memory service that every AI session shares, a gateway that routes to whichever model is cheapest this week, and the monitoring that tells me which of those fell over in the night.
Most posts here start as a ticket on that Git server. Most fixes start with "it worked yesterday".
I'm Lance. I write software for a living and tinker with AI for the rest of it. This site is where the tinkering gets written up: what I tried, what it cost, and whether I'd do it again.
If you're comparing AI models on price, start with the monthly token market. If you're building a homelab, the notes are honest about the bits that broke.