How AI works · who controls the model
Open weights
30-second gist~30s read
An open-weights AI is one whose finished brain — the giant grid of numbers that makes it work — is published for anyone to download. You can run it on your own laptop, fine-tune it, or build a product on top.
The opposite is a closed model, like ChatGPT or Claude, which you can only access through a paid API or app.
If you want more
Why some people care a lot
Three reasons. Privacy — the model runs on your machine, so your data never leaves. Control — you can modify it, censor less, fix bugs, run it offline. Cost — at scale, running a model yourself can be a fraction of paying per API call.
The trade-offs: open-weight models are usually a step or two behind the best closed ones, need real hardware to run quickly, and ship with weaker safety controls. Most ordinary users will never touch one directly — but the apps they use increasingly do.
"Open weights" vs "open source"
These are not the same thing. Open weights means you get the trained model. Open source would mean you also get the training code, training data, and full recipe. Almost no big AI is fully open source — even Meta's Llama and Mistral release weights but keep the training data private.
A real shake-up
In January 2025 a Chinese company called DeepSeek released an open-weights model that came close to matching the top US closed models — at a fraction of the training cost. It briefly knocked hundreds of billions off the value of US tech stocks and reset the conversation about how big a head start the closed labs really have.