French AI Startup Takes Aim at OpenAI and Chinese Rivals
Mistral has launched Mistral Large 4, a trillion-parameter multimodal AI model designed to compete with both closed American systems and increasingly influential Chinese open models. The French company is positioning the model as a European alternative that combines advanced capabilities with the long-term availability promised by open-weight releases. For now, the model is accessible through a controlled endpoint, with public weights expected within three weeks after additional safety testing. Mistral argues that enterprises and government institutions need AI systems they can fully own and audit, particularly in cybersecurity and other critical sectors. The company says Large 4 was trained entirely on its own infrastructure using around 4,000 Nvidia GPUs, significantly fewer than the resources reportedly used by larger rivals. Mistral believes the model can outperform competitors in specialized areas such as cybersecurity, finance, and chip design, even if broader benchmark comparisons are still pending. The launch also reinforces Mistral's ambition to establish Europe as a credible third force in the global AI race, positioned between U.S. closed models and Chinese open ecosystems.