Mistral Large 4 'Le Chonk' Public Preview: A Trillion-Parameter Multimodal Model with Immediate API Access

Mistral AI has launched a public preview of its new flagship model, Mistral Large 4 (ML4), affectionately nicknamed "Le Chonk." This trillion-parameter, natively multimodal model, featuring 52 billion active parameters and a 1-million-token context window, was made available via API on October 6, 2026. The full open weights are slated for release by the end of October. ML4 was trained entirely on Mistral's own European data centers using 3,800 NVIDIA Grace Blackwell GPUs, supporting over 160 languages, including all official EU languages. This release is significant for practitioners because it offers immediate, hands-on access to a frontier-class model through an API, a notable departure from the typical staggered release of open weights. This allows developers and enterprises to begin prototyping and evaluating ML4's capabilities in areas like advanced coding, agentic workflows, cybersecurity, and multimodal understanding without waiting for the full open-weight download. The model's emphasis on European data sovereignty, with training and inference occurring within the EU, directly addresses a critical need for organizations operating under strict data residency and compliance regulations, particularly within Europe. This move by Mistral aligns with the broader trend in the AI landscape where model providers are increasingly offering flexible deployment options and catering to specific regional and industry compliance requirements. While many leading AI labs are focused on pushing the boundaries of model scale and capability, Mistral is simultaneously emphasizing control and sovereignty, a growing concern for enterprises globally. The ability to access a state-of-the-art model with assurances of European infrastructure and data handling provides a compelling alternative to predominantly US-based offerings. This also reflects a maturing AI market where specialized needs beyond raw performance, such as data governance and vendor independence, are becoming key differentiators. In practice, practitioners should leverage the immediate API access to conduct thorough evaluations of ML4's performance on their specific use cases, especially those involving complex reasoning, multimodal input, or sensitive data. For European companies, this model presents a strong candidate for applications requiring strict adherence to GDPR and other regional regulations. Developers should also prepare for the eventual open-weight release by assessing their infrastructure's capability to self-host a trillion-parameter model, which will present a different set of operational challenges compared to smaller models. The focus on cybersecurity performance also suggests ML4 could be a valuable tool for security-sensitive applications, warranting early exploration by security teams.