Open sourceorOpen weight
Open source is a licensing regime: it grants the right to use, study, modify and redistribute, with no discrimination as to use. Open weight states only a fact: the parameters can be downloaded, under conditions that still have to be read.
Point by point
| Criterion | Floor 1Floor 1 · The Modela solid block on its own: the prediction machineOpen source | Floor 1Floor 1 · The Modela solid block on its own: the prediction machineOpen weight |
|---|---|---|
| What the term describes | Rights granted | Availability |
| Restrictions on use | None, by definition | Frequent: thresholds, excluded fields |
| Training data | Required by the recent definition | Almost never published |
| Reproducible | That is the point | No: you can run it without being able to remake it |
| What can be checked | The whole recipe | The behaviour, by measurement |
| Wrongly used for | Most models announced as “open source” |
On the ground four situations
A model released under a recognised permissive licence, with no restriction on commercial use.
Floor 1 · The Modela solid block on its own: the prediction machineOpen sourceThe freedoms are genuinely granted. It remains that the data is not published, which the recent definition of open source in AI would require.
A downloadable model whose licence forbids use beyond a certain number of monthly users.
Floor 1 · The Modela solid block on its own: the prediction machineOpen weightA threshold is a discrimination as to use: the licence is not free, whatever wording the announcement uses.
A model reachable only through the service of its publisher, with no download possible.
Neither one: a closed modelWith no release of the parameters, neither term applies. This is the third category, and it is still the one holding the best-performing models.
A model under a permissive licence, whose training code and corpus composition are published.
BothThe rare case where both labels are right: the weights are open, and the freedoms are granted along with what is needed to rebuild.
The test that settles it
Look in the licence for a restriction on use: a forbidden field, a user threshold, a requirement to seek permission. If there is one, it is not open source, whatever word the announcement uses.
The trap
The confusion is not merely lexical, it is legal. Believing a model is free because you were able to download it exposes you to discovering, once the product has shipped, that a use was forbidden or subject to a commercial agreement.
Check click your answer
Level 1 · Recognise
What does the fact that a model can be downloaded guarantee?
Level 2 · Distinguish
Why does the definition of open source, applied to AI, require information about the data?
Level 2 · Distinguish
What do you check first before integrating a model announced as “open source”?