Open Source VS Open Weight

Open Source, Open Weight, but what are we really talking about?

With the emergence of DeepSeek and Mistral AI, the notion of open source has often come up. But confusion persists between these notions, which are not interchangeable.

•  An Open Weight model: shares only its weights (i.e. the model parameters) without necessarily revealing the training data or source code. You can therefore see the final results, but without fully understanding the training process or replicating the model easily.

" It's like receiving a list of ingredients for a recipe, but without the instructions for preparing the dish."

•   Open Source model: is completely transparent. The code, weights, training data and even the entire training process are available to the public. This allows anyone to study, improve or even create new models from this base.

" It means having the complete recipe, with the list of ingredients, step-by-step instructions, and even the origin of the products used.

At LINAGORA, we have chosen to create LUCIE, a truly Open Source AI.
LUCIE is :

• A dataset available on HuggingFace to ensure transparency and trust
• Training on French, English, Spanish, German and Italian content
• A step towards sovereign and ethical AI for Europe