Mistral AI Releases Mistral Large 4 (Le Chonk): A 1.05T Parameter Multimodal MoE Model

Mistral AI has launched Mistral Large 4, internally nicknamed 'Le Chonk,' as a public preview. The model is a Mixture of Experts architecture with 1.05 trillion total parameters, of which 49 billion are active during inference. It supports native image input and offers a context window of up to one million tokens.

The scale of this release positions Mistral among the top tier of open-weights frontier model providers. The combination of multimodal capabilities and an exceptionally long context window makes the model suitable for document-heavy workloads, codebases, and enterprise retrieval tasks that exceed what most current competitors offer.

Training was conducted on 3,800 NVIDIA Grace Blackwell GPUs, underscoring the growing compute requirements for frontier-scale MoE models. This also highlights NVIDIA's latest architecture being adopted for large-scale training runs by European AI labs.

Mistral AI has been steadily expanding its model portfolio since its founding, moving from smaller open-weight releases to increasingly large architectures. Mistral Large 4 represents a significant step up in parameter count and capability, and its public preview allows developers to evaluate performance before a wider rollout.

The release intensifies competition in the open-weights LLM space, where providers are racing to match proprietary models on multimodal understanding and long-context reasoning. If benchmarks hold up, Mistral Large 4 could become a strong option for enterprises seeking alternatives to closed-source frontier models.

This page provides an editorial summary based on publicly available information. It is not a republished article. Use the source link below for the original report.