Mistral AI Launches 1 Trillion Parameter ML4 Model with 4,000 GPUs

Mistral AI has released the Mistral Large 4 (ML4), a multimodal model featuring 1 trillion parameters, trained using 4,000 Nvidia GPUs. This model utilizes 2 to 3 times less GPU power compared to its Chinese competitors, showcasing significant efficiency. Mistral achieved a valuation of €21 billion (approximately $24.39 billion) during its Series D funding, led by Samsung.

Pierre Stock, Mistral's VP of Science, emphasized that the ML4 model is designed to outperform closed models in specific applications crucial to its customers, such as cybersecurity, finance, and chip design. The model's open-weight architecture is intended to facilitate easier audits, which Stock claims is essential for responsible use. Mistral plans to make the ML4 weights available in three weeks after undergoing safety testing.

This follows Mistral's previous advancements in AI models, positioning the company as a strong contender in the competitive landscape of AI development.