On October 3, Mistral AI introduced its latest large multimodal model, Mistral Large 4 (ML4), aiming to serve as a strong alternative to both closed and open AI models. This release aligns with the vision expressed by French President Macron, who spoke of a “third way in AI” for Europe, emphasizing the need to keep up with advancements from the U.S. and China. Nicknamed Le Chonk, ML4 features an impressive one trillion parameters, positioning it as a formidable player in the AI landscape.
At this stage, ML4 is not an open-weight model and is only accessible through a public guardrail endpoint. Mistral plans to release the model weights in three weeks, following the completion of safety testing. Mistral’s VP of Science, Pierre Stock, said the company will work with trusted partners and governments to ensure the open-source weights can be used to defend against threats but not to perform malicious attacks. This strategy addresses growing security concerns, particularly among Mistral’s main users–enterprises and institutions. Stock also mentioned that an open-weight model is easier to audit.
One notable aspect of ML4’s development is that it was trained entirely on Mistral’s own compute infrastructure, using just 4,000 NVIDIA GPUs–two to three times fewer than many Chinese models and significantly less than closed-source alternatives. Stock said Mistral hopes ML4 will be best in class among open-weight models, especially outside of China, and could also outperform closed models in specific areas key to its customers. These areas include cybersecurity, finance, and chip design, the latter being critical to two of Mistral’s major investors: ASML and Samsung, who recently valued the company at €21 billion (approximately $24.39 billion).
While Mistral plans to host models from China, it insists this won’t reduce its role to merely an inference provider. With Le Chonk in its corner, Mistral believes it should still be considered a frontier lab, not merely an inference provider.
