Mistral’s new 1T model aims to leapfrog closed and open rivals
techcrunch ·
French AI lab Mistral AI has released Mistral Large 4, a new multimodal model with one trillion parameters designed to compete with both American and Chinese rivals. Nicknamed 'Le Chonk,' the model is positioned as part of Europe's 'third way' in artificial intelligence, offering an alternative to closed and open-source systems. Although not yet available as an open-weight model, Mistral plans to release the weights after three weeks of safety testing to ensure they can be used for defense rather than malicious attacks. The model was trained entirely on Mistral’s own infrastructure using only 4,000 NVIDIA GPUs, which is significantly less compute than its competitors. Key use cases include cybersecurity, finance, and chip design, supported by major backers like ASML and Samsung. While benchmark results are pending, Mistral aims to lead among open-weight models outside China while maintaining its status as a frontier laboratory.
techcrunch