Open weights vs open source AI: what the OSI definition requires

Open weights means you can download the trained parameters. Open source AI, as the Open Source Initiative defines it, asks for much more: the freedoms to use, study, modify and share, plus data information, training code and parameters under approved terms. Many popular local models meet the first bar and not the second.

What the definition says

The Open Source AI Definition, version 1.0, lists four freedoms. Users must be able to use the system for any purpose without asking permission, study how it works and inspect its components, modify it for any purpose including changing its output, and share it with or without modifications for any purpose.

To make modification practical, it names the preferred form for making changes. That has three parts:

1. Data information. The definition asks for sufficiently detailed information about the data used to train the system. That includes a description of all training data, including data that cannot be shared, plus lists of publicly available and third party data and where to get them.
2. Code. The complete source code used to train and run the system, including data processing, training settings, validation, supporting libraries, inference and model architecture, under OSI approved licenses.
3. Parameters. The model parameters, such as weights or other configuration settings, under OSI approved terms.

What that means for the model you downloaded

A typical open weights release gives you the weights and inference code. Whether it also gives training code and detailed data information varies by release. Whether the license permits any purpose varies too: some licenses add use restrictions or conditions for large deployments, which the four freedoms would not allow.

So the useful question is not whether a model is open. Ask which of these you actually have:

• the weights
• the inference code
• the training code
• information about the training data
• a license that lets you use, modify and redistribute without extra conditions

Why developers should care

For a developer running a model locally, the weights are what make private use possible. You can run, test, quantize and fine tune what you have.

The gaps matter in other ways:

• Reproducibility. Without training code and data information you cannot recreate or audit the model.
• Legal clarity. A custom license may limit commercial use, redistribution of fine tunes, or use in certain products. Read it before you build a product on it.
• Supply chain review. The more you can inspect, the more a security team can reason about the model. With weights only, you are trusting the publisher more.

A practical approach for teams

Keep an inventory of the models in use with their license text, source URL and file hash. Have legal review any license that is not a standard OSI approved one. Do not describe a model as open source in your own materials unless it meets the definition, because the term now has a published meaning.

For the Apache 2.0 case see the Apache 2.0 redistribution checklist. For deployment risk see local LLM deployment risks.

Quick check

Sources

• The Open Source AI Definition 1.0, Open Source Initiative

Last reviewed: 19 October 2026.