LANDMARK · 7 MIN · UNDERSTAND
Open and closed models
After this landmark, you can read the access and licence terms of a model accurately, and choose between hosted and self-hosted on the grounds that actually differ.
↗
This concept is shared, but the Everyday lens is less central here.
You can still explore it. We’re showing the shared explanation and a related practical view without hiding the knowledge.
Models sit on a spectrum of access. Closed models are reachable only through an API: you send text, you get text, the weights stay with the provider. Open-weight models can be downloaded and run on your own hardware. A genuinely open-source model would also publish its training data and code, which is rare — most “open” models publish weights under a licence with conditions, and the label is doing a lot of work it hasn’t earned. The distinction matters for four practical reasons. Data: with self-hosting, prompts never leave your infrastructure. Stability: a downloaded model doesn’t change under you, while a hosted one can be updated or retired on the provider’s schedule. Cost shape: hosted is per-token operating expense, self-hosted is hardware and staff. And control: only weights you hold can be fine-tuned freely, inspected, or kept running after a vendor decides otherwise.
This is the canonical concept. It stays the same across learner lenses so personalization never changes the underlying facts.
◌ EVERYDAY LENS
What this looks like for you
Local models on a laptop or phone are open-weight models. They are weaker at hard problems and they keep your data on the device, which for personal or sensitive material is frequently the better trade.
MAKE A DECISION
A regulated team must process confidential case files with AI and reproduce any decision on request for the next seven years. What matters most in the choice?
CARRY THISFor one model your team uses, write down: can you download the weights, what does the licence restrict, and could you still run this exact version in three years?