Open-Weight Model
Also called Open weights, Open-weights model, Downloadable model
An AI model whose trained weights are published for anyone to download and run, though its training data and code may stay private.
What is an open-weight model?
An open-weight model is an AI model whose weights, the billions of numbers learned during training, are released publicly. You can download the model, run it on your own hardware, and usually fine-tune it, instead of only reaching it through a company's app or API.
Well-known families of open-weight models come from Meta, Mistral, Google, Alibaba, DeepSeek and others.
How does it work?
The developer publishes the model files, typically on a site such as Hugging Face, along with a licence. You then run them with local apps or on your own servers. Most people use quantized versions, which are much smaller and run on ordinary computers with only a small loss in quality.
What is it commonly confused with?
- Open-source AI. By the Open Source Initiative's definition, open-source AI also needs enough information about the training data, and the code used to train and run the model, for others to study and rebuild it. Many open-weight models publish only the weights, so they're open-weight but not fully open source.
- Free to use for anything. Licences differ. Some are permissive; others restrict commercial use, large companies or certain uses. Read the licence before you build on a model.
Why does it matter?
Open weights give you control: your data stays on your hardware, nobody can change or retire the model you depend on, and there's no per-request fee once you have the hardware. The trade-offs are running and securing it yourself, and the most capable models are often still closed. Read Open-source vs closed AI models for the full comparison.
