Open-weight vs closed AI models
Updated · 6 min read
The AI behind a chatbot is either closed — available only through its maker — or open-weight, published so anyone can run it. The difference shapes privacy, cost, reliability and control.
01
What “open-weight” means
A model’s weights are the billions of numbers it learned during training. When they are published, anyone with the hardware can run the model — a cloud provider, a Canadian data centre, or your own server. Closed models are reachable only through their maker’s service.
“Open-weight” is not always the same as “open source”: some models publish weights but not training data or code, and licences vary. For running a model where you choose, the weights are what matter.
02
How they compare
- 01
Quality
the best open-weight models are now competitive with leading closed models on many everyday and expert tasks.
- 02
Control
open models can be hosted where you choose; closed models run where their vendor runs them.
- 03
Lock-in
switching between open models is routine; leaving a closed ecosystem can be hard.
- 04
Cost
open models can be served by competing providers, which tends to push prices down.
- 05
Privacy
with open weights, the company hosting the model can be one you trust, in a jurisdiction you choose.
03
Where closed models still lead
On some of the hardest benchmarks and newest capabilities, the largest closed models often arrive first. The gap has been narrowing quickly, and for most everyday work — writing, summarising, research with sources, coding help — well-chosen open models are more than capable.
04
Why not just pick one?
Different models are good at different things, and every model is sometimes confidently wrong. Using several — routing each question to a good fit and letting one model check another — gives better answers than any single model alone.
Frequently asked questions
Are open-weight models safe to use?
The safety of a model depends on how it is tuned and deployed. Hosting an open model yourself or with a trusted provider also means your prompts go only where you choose.
Can I run an open model on my laptop?
Smaller ones, yes. The strongest open models need data-centre GPUs, which is why Canadian hosting matters.
Try asking Digby
Keep reading
Try Digby now.
Free, no account needed. Models and data stay in Canada.