Named "Croissant LLM", the main features of this model are:

- Sovereign: trained on the Jean Zay calculator with open data
- Accountable: fully sourced data for total transparency
- Ethical: advanced compliance with AI Act regulations
- Frugality and speed: runs on CPU and phone because it's so compact
- Benchmark: stands out as the best-performing French-language model for its size
- Commercial use: possible for both data and model
- Integrates French cultural specificities for an enriched model.

This innovation was developed by Professors Pierre Colombo and Céline Hudelot as part of Manuel Faysse's thesis work, in collaboration with Nuno Miguel Guerreiro and Patrick Fernandes. This work is the fruit of close collaboration between academia and industry, illustrating the importance of synergy in advancing AI research.

CroissantLLM is the result of an association between CentraleSupélec and several renowned academic partners such as Sorbonne Université, INESC-ID, Instituto Superior Técnico, Carnegie Mellon University and Institut DATAIA, and the invaluable support of industrial partners such as ILLUIN Technology, Unbabel, Diabolocom, et EqualAI.