Mistral releases Mistral Large 4, dubbed "le Chonk", a 1T-parameter open-weight model for general agentic capabilities, trained on 4,000 Grace Blackwell GPUs (Sabrina Ortiz/The Deep View)
Most flagship models dominating the market are from frontier labs and are closed models. French AI lab Mistral seeks to change that with its latest model.
French AI startup Mistral has unveiled its latest model, Mistral Large 4 (ML4), positioning it as an open solution to cater to various security needs. The company emphasizes that open models are becoming a viable option for cybersecurity, as opposed to proprietary models from industry giants like Anthropic and OpenAI, which have been proven to be equally risky.
ML4, nicknamed "Le Chonk" due to its immense size of a trillion parameters, is designed to offer robust security measures that remain under the user's control. Unlike closed models, Mistral's open model cannot be withdrawn by the developers, ensuring enterprises and governments always have access to defend against cyber threats.
Mistral's co-founder, Guillaume Lample, highlighted that ML4 is tailored for cyber defense, enabling users to maintain control over data and ensure security. Both Mistral's CEO, Guillaume Lample, and VP of Science, Pierre Stock, stressed the significance of security for enterprise clients, reflecting the growing trend in the industry. The new model comes in response to a series of AI-related hacks that have underscored the need for dependable security solutions.
In a briefing, Lample and Stock emphasized that security is a crucial factor for Mistral's enterprise clients, echoing the industry's shift towards prioritizing cybersecurity. Following the Hugging Face breach, Mistral joined Nvidia's Open Secure AI Alliance, an alliance aiming to democratize cyber defenses by promoting open models.
The company claims that ML4 leads the generation of open-weight, customizable cybersecurity models, allowing enterprises to maintain full ownership and control over their AI systems without being susceptible to vendor lock-in.
Mistral's ML4 model has proven to be competitive with pricier proprietary models, such as GPT-6 Astra, and even open-weight Chinese models like Kimi K3. It achieved a 15% improvement over open-weight models on Harvey's Legal Agent benchmark, indicating its strong performance in open-weight models. The availability of ML4 in a public preview format and its compatibility with Nvidia's Grace Blackwell GPUs demonstrate Mistral's commitment to providing an accessible and high-performing model for various industries.
Written by urgent.news from ZDNet'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.
