In short
- Open-weight means a model's trained parameters, its weights, are published so you can download and run it.
- Open-source AI, as the Open Source Initiative defines it, also needs the training code and enough information about the training data to build a substantially equivalent system.
- Most “open” models are open-weight, and their licenses can still limit how you use them.
- Before you build on one, read the license, check what is actually published, and pin the exact version.
What is an open-weight model?
A trained AI model is, in the end, a very large set of numbers called parameters or weights. They hold what the model learned. An open-weight release publishes those numbers, usually as files you can download, so you can run the model on your own hardware, fine-tune it, or inspect how it behaves.
That is all the term promises. It says nothing, on its own, about the code used to train the model, the data it was trained on, or what you are legally allowed to do with it. Those depend on what else is published and on the license.
What does open-source AI require?
The Open Source Initiative, which maintains the Open Source Definition for software, published version 1.0 of its Open Source AI Definition in October 2024. It says an open-source AI system must give everyone the freedom to use it for any purpose, to study how it works, to modify it for any purpose, and to share it, with or without modifications.
To make those freedoms real, the definition asks for the “preferred form” for making changes, which has three parts: sufficiently detailed information about the training data, the complete code used to train and run the system, and the model parameters. On that definition, a release of weights alone, or weights under a restrictive license, is not open-source AI.
What the difference means in practice
| Question | Open-weight release | Open-source AI (OSI definition) |
|---|---|---|
| Run it on your own hardware | Yes | Yes |
| Test how it behaves | Yes | Yes |
| Fine-tune it | Usually, if the license allows | Yes |
| Learn what data trained it | Only if the publisher says | Detailed information is required |
| Retrain it from scratch | No, the code and data details may be missing | In principle, yes |
| Use it for any purpose | Depends entirely on the license | Yes, that is a core freedom |
License terms vary widely. Some well-known open-weight models ship under custom licenses that limit certain uses, for example by field of use or by the size of the company using the model. Others use permissive licenses. The word “open” on a model page tells you little. The license text tells you what you can do.
What open weights give you
- Control of your data. You can run the model on your own infrastructure, so inputs and outputs don't have to leave it.
- Auditability. Researchers and security teams can test behavior directly rather than relying on a description.
- Stability. A model you have downloaded doesn't change underneath you, so results can be reproduced later.
- Customization. You can fine-tune a model or adapt it to your hardware.
- Offline use. Nothing needs to phone home.
What open weights don't give you
Open weights don't tell you what the model was trained on, don't let you recreate the model, and don't guarantee quality, safety or freedom from license terms. They also don't make a model automatically better for your task. You still have to test it on your own data.
A checklist before you build on an open model
- Read the license. Look for limits on commercial use, redistribution, field of use and attribution.
- Check what is published. Weights only, or also the training code and a description of the data?
- Find the model card. It should state intended uses, known limits and how the model was evaluated.
- Pin the version. Record the exact release and the checksum of the files you use.
- Test on your own data. Published results rarely match your inputs.
- Plan for updates. Decide how you will hear about security fixes or license changes.
How we use the terms
We call Synas Image 001 and Synas Detect 001 open-weight, not open-source, on purpose. It is the accurate word for a release of weights, and each model's license will be published with its release so you can judge for yourself.
Sources
- The Open Source Initiative: The Open Source AI Definition, version 1.0. Published October 2024