How the series measures things
Every comparison in the series follows the same disciplined routine, so the only thing moving between tests is the thing being tested.
Same prompt
The exact same prompt is used across the board, so every model is judged on identical input.
Same rig
Every run happens on the same machine — one fixed setup, so hardware is never the variable.
One variable per test
Each test changes exactly one variable. That keeps the outcome attributable, not muddled.
Model reloaded before every run
The model is reloaded before every single run, so no test inherits state from the one before it.
Everything published
Every prompt and every result file is published in the open — nothing stays behind the camera.
Repeatable by design. Because the prompt, the rig, and the procedure are fixed and every prompt and result file is published, anyone can reproduce a test from the repository.
One rig, no exceptions
The whole series runs on a single machine. Here is exactly what is doing the work.
The rig
Run by PC Magik — Mateusz Piekut
PC Magik Lab is made by PC Magik (Mateusz Piekut), a computer service and homelab operator from Poland. The series puts AI models to work on the kind of real, local hardware he lives and works with every day.
No rented clusters, no hand-waving — just local models, real GPUs, and a method you can follow yourself.
Follow the experiments
Watch the series on YouTube, or dig into the repository where every prompt and result file is published.