Mistral Large 4

Mistral Large 4 is a general-purpose multimodal model built on a Mixture-of-Experts architecture, with 52 billion active parameters, 1.05 trillion total parameters and a 1.6-billion-parameter vision encoder. Its documented context window is one million tokens. The model is in public preview through Mistral’s API, with open weights planned for release later in October 2026. The model supports structured outputs, function calling, document question answering, chat completions, batching and agent or conversation workflows. Its multimodal capabilities are intended for work involving documents, charts and images, while its general-purpose use cases include coding, reasoning and tool-based tasks. These capabilities make it relevant to developers building assistants and agents, and to teams handling long documents or visual information. Mistral publishes separate API rates for input, cached input and output tokens. That pricing lets developers estimate costs by prompt and response volume, with cached input billed at a lower rate. Organizations interested in running the model under their own infrastructure can follow the announced open-weight release, while API users can try the preview without managing model weights.