Mistral Large 4 is a Large Language Models (LLMs) tool. Open-weight multimodal AI model with 1 trillion parameters for coding and reasoning tasks. Key features include Mixture of Experts Architecture, Multimodal Input Processing, and Extended Context Window. Best for software developers and engineers, data scientists and analysts and scientists and researchers.
About Mistral Large 4
Mistral Large 4 is a frontier multimodal large language model with 1 trillion parameters and 49 billion active parameters. It handles text and image inputs, excels at coding, cybersecurity, and agentic workflows, and supports over 160 languages.
Key Features
Mixture of Experts Architecture.
Multimodal Input Processing.
Extended Context Window.
Advanced Coding Capabilities.
Cybersecurity Strengths.
Multilingual Support.
Frequently Asked Questions
Mistral Large 4 is a frontier large language model from Mistral AI with 1 trillion parameters. It uses a mixture of experts architecture that activates 49 billion parameters per request. The model handles text and image inputs and excels at coding, reasoning, cybersecurity, and agentic workflows.
Mistral Large 4 costs $0.68 per million input tokens and $2.09 per million output tokens through the API. The pricing is currently discounted by 50% during the public preview period. Cached input costs $0.07 per million tokens.
Mistral Large 4 weights are scheduled for public release by the end of October 2026. The model launched in public preview on October 6, 2026, with the open weights following later. Before release, Mistral is conducting safety testing with cybersecurity leaders and government authorities.
Mistral Large 4 stands out for its strong performance in specialized domains like cybersecurity, coding, and computer vision. It outperformed GPT-6 Astra on certain vision benchmarks and scored ahead of open-source rivals like Qwen and DeepSeek on coding and knowledge work tasks. It also offers European data sovereignty options.
