Local AI Benchmarks

Reproducible benchmarks of open-weights language models running on consumer hardware. Generation speed, repetition rates, memory footprints, and cost-per-book measured across three machines spanning six years of GPU generations — with the scripts, raw outputs, and failed runs published in full.

Is Bigger Better? Thirty-Seven AI Novels Say No

Is Bigger Better? Thirty-Seven AI Novels Say No
I generated the same 90,000-word novel 37 times across eight local AI models and three graphics cards, to find out which combination actually works on hardware you already own. Both halves of "bigger is better" turned out to be wrong — and the biggest improvement I fo...
Read More

Can AI Write a Book?

Can AI Write a Book?
Can a local AI model actually write a novel—and can it do it on hardware normal people already own? I’m benchmarking AI book generation across older GPUs, measuring speed, repetition, memory use, editing burden, and the surprisingly tiny cost of generating a 90,000-...
Read More