AI2 min
French AI lab Mistral publishes the weights of Mistral Large 4, its largest model
Mistral has released the weights of Mistral Large 4, which it also calls Le Chonk. Weights are the numbers that make up a model, so anyone can download and run it. Mistral says it is a 1 trillion-parameter mixture-of-experts model. Only about 49 billion parameters are used for each token, because each token goes to just 4 of 128 expert sub-networks. It reads text and images, writes text, handles a context of up to 1 million tokens and has three reasoning levels: low, medium and high. Mistral says it trained the model from scratch on 3,800 Nvidia Grace Blackwell GPUs in its own European data centres, and that it works in more than 160 languages. Mistral says it does especially well on cybersecurity, finance and law work, and TechCrunch notes the company is pitching it to organisations that want control over their cyber defences. It is also available through Mistral's own API.
Why it matters: This is one of the largest models anyone can download. Running it yourself still takes serious server hardware, so most people will use it through Mistral's API or cloud providers. The performance claims come from Mistral.