Mistral says ML4 was trained using 3,800 Nvidia Grace Blackwell GPUs in its own data centers in Europe and much of its training data was multilingual (Mistral Blog)
Mistral is framing ML4 as a European-built flagship, trained on its own compute and prepared for open-weight release.
The Techmeme note points to Mistral’s public preview of Mistral Large 4, nicknamed “le Chonk.” It says the model was trained on 3,800 Nvidia Grace Blackwell GPUs in Mistral’s European data centers. The item also highlights Mistral’s claim that much of the training data was multilingual. Source: Techmeme's note
The Techmeme note points to Mistral’s public preview of Mistral Large 4, nicknamed “le Chonk.” It says the model was trained on 3,800 Nvidia Grace Blackwell GPUs in Mistral’s European data centers. The item also highlights Mistral’s claim that much of the training data was multilingual. Source: Techmeme's note
score 8