For a long time the best models lived behind an API, and that was that. The gap between what you could download and run yourself and what the big labs offered felt permanent. It isn’t anymore.
Open-weight models — ones whose parameters you can download and run on your own hardware — have been climbing fast. The newest releases handle reasoning, coding, and long documents at a level that would have been frontier-grade not long ago. The headline isn’t that they beat the very best closed models; it’s that for a lot of real work, the difference no longer matters.
Why this matters
Running a capable model yourself changes the math. Your data never leaves your machine, costs become predictable, and you’re not exposed to a vendor changing prices or pulling a model you depend on. For privacy-sensitive work, that control is the whole point.
The catch
Self-hosting isn’t free in the ways that matter to most people: you need the hardware, the setup, and someone to keep it running. For a solo user, a hosted model is still simpler. But the option now exists, and it’s good enough that ignoring it would be a mistake for any team handling sensitive information.