Mistral Large 4: "Le Chonk"
Article URL: https://mistral.ai/news/mistral-large-4/ Comments URL: https://news.ycombinator.com/item?id=49978116 Points: 250 # Comments: 28
Mistral AI has announced the public preview of its latest model, Mistral Large 4, also unofficially referred to as ML4 or le Chonk. This 1 trillion-parameter, natively multimodal model boasts 49 billion active parameters, marking it as the company's largest and most capable model to date.
ML4 demonstrates exceptional performance across various domains, including coding, agentic workflows, and multimodal understanding. It already performs as well as the strongest open-source models globally while significantly outperforming open-weight models developed in the US or Europe. In critical enterprise workloads such as cybersecurity, finance, and law, ML4 has been found to be state-of-the-art among open models.
The model was trained from scratch on 3,800 NVIDIA Grace Blackwell GPUs housed in Mistral's European datacenters. The public preview is currently served on the same infrastructure. Companies can access ML4 via the Mistral Studio and can look forward to the weights being released by the end of the month. In the meantime, the model is being red-teamed in real-world settings with cybersecurity leaders, vetted partners, and state authorities.
Mistral AI has invested significantly in infrastructure, research, and product development to deliver state-of-the-art performance in critical verticals through open weights. This is particularly important in sectors like cybersecurity, where provider-level refusals can block legitimate vulnerability research and incident response. With ML4, organizations gain control over their AI while benefiting from top-tier cyber performance.
ML4 was trained on multilingual data spanning over 160 languages, including every official language of the European Union. The model was trained using the same environment offered to customers through Mistral Forge. As the company awaits the release of the weights, more details on the model architecture, additional benchmarks, and post-training methodology will be shared.
Written by urgent.news from Hacker News Best's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
This story
This is one outlet's version. Read the fullest account.
- Mistral Large 4: "Le Chonk" mistral.ai