Glossary · 1 · Foundations: how AI works
Open-weight model
Also known as: Open model, Small language model (SLM), Self-hosted model
An open-weight model is an AI model whose trained weights are published for download, so organizations can run, inspect and adapt it on their own infrastructure — under a license that may or may not meet the definition of open source.
- Intermediate
- Technical marketers
- Technical project managers
- Developers
In one sentence
Open-weight models explained: downloadable AI models you can run yourself — benefits, licenses, and the difference from open source.
Example
A manufacturer runs an open-weight model on its own servers so that confidential service documentation never leaves the company network.
Why it matters on your learning path
- Technical marketers: “Open” is a loaded word. Say open-weight unless the license and training data meet the Open Source Initiative’s definition.
- Technical project managers: Self-hosting gives data control and cost predictability but adds operations, security and update work.
- Developers: Families such as Llama, Mistral, Qwen, DeepSeek, Gemma and Granite are available on Hugging Face; small models run on laptops and devices.
Open-weight vs. closed vs. open source
Closed models are only reachable through the vendor’s app or API. Open-weight models publish weights, often with use restrictions. Open-source AI, as defined by the Open Source Initiative (OSAID 1.0), additionally requires enough information about training data and code to rebuild the model.