An open-weight model is an AI model whose weights are published. Anyone can download a copy and run it on their own computers, though it often comes with a license that sets conditions on how you may use it. Open-weight does not always mean open source, because the company that made it may keep its training data and code private. For example, a small food bank could download an open model and run it on a computer in its own office instead of typing questions into a chatbot website.
Why it matters
Running an open-weight model in-house keeps your prompts on hardware you control rather than sending them to an outside service. That can matter for sensitive notes about clients or donors. The trade-off is that you take on work a vendor would normally do. Someone has to keep the software up to date, patch the inference server that runs the model, and secure the machine it sits on. Before you choose one, read the license terms and decide who on your staff or among your IT helpers will own those updates. If a vendor says its product uses an open model, ask whether the model runs on your hardware or on theirs.
Published Updated