Mistral Large 4 (Le Chonk): 1 Trillion-Parameter Open Model Released

What changed?

Mistral Large 4 (Le Chonk) is now available as the company's most capable open-weight model to date. It is a 1 trillion-parameter, natively multimodal Mixture-of-Experts model supporting 49 billion active parameters per token (52 billion including embeddings and output layers). The checkpoint supports both text and image input, with a 1 million token context window. The release also introduces new quantization features, improved memory efficiency during training, and bug fixes. The model is available under the Mistral Ecosystem Open-Weight License, a modified MIT license permitting broad use, except for operating a model-as-a-service business without a separate agreement.

Why does it matter to an everyday developer?

Mistral Large 4 enables developers to build applications with major improvements in multilingual capabilities, visual understanding, and long-context processing. Its context window supports up to 1M tokens, which is significantly higher than most previous models, making it suitable for working with long documents or complex multi-turn workflows. The model supports both text and image input and has advanced reasoning and system prompt adherence, opening new agentic and automation use cases (including native function calling and strict JSON output). Notably, the model performs particularly well on cybersecurity, finance, and law workloads, and offers state-of-the-art open-weight visual grounding. Being open-weight under a broadly permissive license, it enables commercial product development and on-prem deployment.

What can the developer do now?

• Download and use Mistral Large 4 for multimodal (text and image) applications, including agentic workflows, coding assistants, and visual QA.
• Explore long-context use cases such as document analysis, legal reasoning, or technical workflows leveraging the 1 million token window.