Mistral AI has unveiled its latest multimodal model, Mistral Large 4, which features one trillion parameters. The company intends to release the model weights in three weeks, following the completion of safety testing. Currently, the model is accessible exclusively through a public guardrail endpoint.
The development of the model involved training on 4,000 NVIDIA GPUs. According to the company, this infrastructure requirement is significantly lower than that of its closed-source competitors and two to three times less than that of its Chinese rivals.
Mistral AI is targeting specific sectors for the model, including cybersecurity, finance, and chip design. These areas align with the strategic interests of the company's key investors, such as ASML and Samsung.
Samsung led the company's Series D funding round last month, which resulted in a valuation of €21 billion, or approximately $24.39 billion. Mistral AI maintains that it remains a frontier laboratory despite its recent expansion into hosting third-party models.
