aiopensourceservices

Fact sheet · updated

WHAT OPEN REALLY MEANS

Open weights are
not the whole recipe.

Many AI models are called open.

Check a release
A wooden toolbox standing open on a workshop bench, its hand tools neatly arranged in the morning light
AI-generated image, not a photograph of a real person or place.

OPEN?check what ships

Some let you download the trained weights and little else. The Open Source AI Definition asks for more: the weights, the complete code, and enough information about the training data for a skilled person to build a substantially equivalent system.

At a glance

17 parts

Parts of a model release it tracks

The framework sorts 17 components, from architecture and weights to datasets and logs, into three classes: Open Model, Open Tooling and Open Science.

COMPONENTS · MODEL OPENNESS FRAMEWORK

64.67%

Models on Hugging Face Hub with no license

The framework's authors cite research finding that 64.67% of models and 72.13% of datasets on the Hugging Face Hub were shared without a license.

LICENSES · MODEL OPENNESS FRAMEWORK

TWO NUMBERS WORTH KNOWING

A model is more
than its weights.

One figure on how much a full release contains,
one on how often licenses are missing.

The authors say models whose weights are simply published for download are sometimes promoted as open source when they may more accurately be called open-weight models.

A QUICK RELEASE CHECK

What did they
actually release?

Pick what a model's publisher has shared to see how far it goes toward the Open Source AI Definition. The bar shows how complete the release is.

Based on the OSI Open Source AI Definition 1.0 and the Model Openness Framework. General information only, not legal advice about any license.

Release check

Choose what was shared
Access through an online service onlyNot open

You can use the model but cannot download its parameters or inspect its code, so the freedoms to study and modify it are missing.

Weights to download, with use restrictionsOpen weights

Downloadable weights are useful, but the Model Openness Framework does not count a release as open if its license adds restrictions or acceptable-use conditions.

Weights, architecture and docs, openly licensedOpen model

This matches the framework's Class III: architecture, final weights, a technical report, evaluation results, and model and data cards under open licenses.

Plus training, evaluation and inference codeOpen tooling

Class II adds the code to train, test and run the model, so others can check the publisher's claims. The datasets themselves are still not included.

Plus datasets, data code and checkpointsOpen science

Class I adds the research paper, datasets, data preprocessing code and intermediate checkpoints: the gold standard for reproducing the work.

Sources: OSI Open Source AI Definition 1.0 (2024); White et al., The Model Openness Framework, arXiv:2403.13784 (Table 1 and section 4).

THE SHORT ANSWER

Open source AI
means three things.

The Open Source AI Definition 1.0 was released on October 28, 2024.
It lists what has to be shared, and under what terms.

WEIGHTS

The trained parameters

The model's weights and other settings, under OSI-approved terms. The definition gives intermediate checkpoints and the final optimizer state as examples.

Check a release

CODE

The complete source code

The code used to train and run the system, including data processing and filtering, training settings, testing and inference, under OSI-approved licenses.

What to look for

DATA

Data information

Enough detail about the training data that a skilled person can build a substantially equivalent system, including where public and third-party data can be obtained.

See the numbers
Rows of plain cardboard archive boxes on tall metal shelves in a quiet records room
AI-generated image, not a photograph of a real person or place.

FOUR THINGS TO CHECK

Read the release,
not the label.

Before relying on a model described as open, look at what was actually published and the terms attached to each part.

  1. Find the license for each part

    Weights, code and data can each carry different terms. A missing license is not the same as an open one.

  2. Look for use restrictions

    Acceptable-use clauses or limits on commercial use mean the release does not meet the open source freedoms to use it for any purpose.

  3. Check for training code

    Inference code lets you run a model. Training, evaluation and data processing code let you check and rebuild it.

  4. Ask what the data was

    The definition expects a full description of the training data, its provenance and how it was selected and filtered.

WHY IT MATTERS

Open models are
on the US agenda.

America's AI Action Plan, released by the White House on July 23, 2025, has a section on encouraging open-source and open-weight AI. It says startups can use open models without being dependent on a closed model provider, and that many businesses and governments have sensitive data they cannot send to closed model vendors.

The plan asks the Commerce Department's NTIA to convene stakeholders to help small and medium-sized businesses adopt open-source and open-weight models. A year earlier, in July 2024, NTIA had recommended monitoring the risks of openly available model weights rather than immediately restricting them.

An old brass balance scale on a desk with a row of small brass weights beside it
AI-generated image, not a photograph of a real person or place.

Questions

Questions people ask

What is the difference between open source AI and open weights?

Open weights means the trained parameters can be downloaded. The Open Source AI Definition also requires the complete training and running code and detailed information about the training data, all under OSI-approved terms.

Does open source AI have to publish all its training data?

Not always. The definition asks for sufficiently detailed data information, including descriptions of unshareable data, plus listings of public and third-party data and where to obtain it, so a skilled person can build a substantially equivalent system.

Who wrote the Open Source AI Definition?

The Open Source Initiative released version 1.0 on October 28, 2024 after a year-long, global co-design process. Its four freedoms, to use, study, modify and share, are derived from the Free Software Definition.

What is the Model Openness Framework?

A three-tier classification published by Matt White and colleagues in 2024. It lists 17 components of a model release and asks which are shared under open licenses, from Class III Open Model up to Class I Open Science.

Can a license with usage restrictions still be open source?

Under the Model Openness Framework, no: modified open licenses that add restrictions or acceptable uses do not qualify. The Open Source AI Definition requires the freedom to use the system for any purpose without asking permission.

Sources

Where the facts come from

  1. Open Source Initiative: The Open Source AI Definition 1.0
  2. Open Source Initiative: release of the Open Source AI Definition (October 28, 2024)
  3. White et al. (2024), The Model Openness Framework, arXiv:2403.13784
  4. NTIA: NTIA supports open models to promote AI innovation (July 30, 2024)
  5. The White House (July 23, 2025): Winning the Race, America's AI Action Plan (PDF)
  6. Open Source Initiative (August 12, 2025): White House releases AI Action Plan, includes open source

ABOUT THIS NAME

Interested in leasing aiopensourceservices.com?

aiopensourceservices.com is part of the DoodleSale portfolio. It isn't in an auction right now, but you can ask about leasing it.

Ask about this name