What does "open source" mean for software and AI?
Open source software publishes its source code under a license that lets anyone use, study, change and share it. For AI it is murkier: many "open" models only release their weights (the trained numbers) under restrictive licenses — often called "open weights" — which is not the same as open source.
Open source is a recipe anyone can read, cook and change. "Open weights" is a finished cake you're allowed to take home — but not the recipe, and with rules about who can eat it.
- Closed: use it only through the company's app or API
- Open weights: download and run the model, under the vendor's licence (e.g. Llama)
- Open source: code, weights and training-data details, free to use and change
Positions are categories, not scores.
Open source software
With open source software you can read the code yourself, which is why it matters for trust: privacy and security claims can be checked instead of taken on faith. Examples on PixPause include the Signal messenger and coding tools like Aider, Cline and Continue.
Open source AI vs open weights
An AI model is more than code: it has weights (the parameters learned in training) and the data it was trained on. In October 2024 the Open Source Initiative published the Open Source AI Definition 1.0, which requires the code, the weights and sufficiently detailed information about the training data to be available under terms that let anyone use, study, modify and share the system.
Many popular "open" models do not meet that bar. Meta's Llama license, for example, restricts some uses and requires companies with more than 700 million monthly users to ask Meta for permission; the OSI has said the Llama license is not open source. Releasing downloadable weights is still genuinely useful — you can run the model yourself — but "open weights" is the more accurate label.
Why it matters to you
Open code means outsiders can spot problems — and sometimes a company open-sources an app after being caught, as happened with ZCode. Open weights mean you can run a model offline, on your own device, without sending data to the vendor.
▸ THE HYPE VS WHAT WE FOUND
Real claims involving this term, each checked in a PixProof with linked sources.
▸ PIXPROOFS THAT USE THIS TERM (12)
▸ COMPANIES BEHIND IT
HYPE OR REAL?
"After ZCode was caught uploading whole workspaces, its maker open-sourced the app and removed the upload code."
The answer: Partly. The upload code is gone as of v3.14.0 — but Z.ai's claim that the collected data was deleted can't be independently checked.
See the full ZCode PixProof → Play 12 more in Buy or Bye? →▸ QUICK QUESTIONS
- Is Llama open source?
- Meta calls it open, but its license adds usage restrictions, and the Open Source Initiative says it does not meet the open source definition. "Open weights" is the more precise term.
- What are open weights?
- The trained parameters of an AI model, published so anyone can download and run the model — without necessarily sharing the training code, data, or open-source-style usage rights.
▸ SOURCES
- Open Source Initiative — The Open Source AI Definition 1.0 ↗
- Open Source Initiative — Meta's LLaMa license is still not Open Source ↗
- Meta Llama 3 Community License (GitHub) ↗
- Wikipedia — Open-source artificial intelligence ↗
Spot something wrong? Tell us — corrections are logged, see methodology.