Mistral Large 3 is a flagship large language model developed by Mistral AI, officially announced in February 2025. It is a 123 billion parameter dense model designed to deliver high-tier reasoning, advanced coding capabilities, and extensive multilingual support. As the successor to the Mistral Large 2 series, it is engineered for high-complexity enterprise tasks and the development of sophisticated agentic workflows.
The model features a 128,000 token context window, enabling the processing of extensive technical documentation and complex, multi-turn conversational histories. Mistral Large 3 has been optimized for precision in instruction following and tool calling, positioning it as a tool for automated systems and synthetic data generation. Its performance highlights include strong results on benchmarks for mathematics, logic, and programming across various languages.
In conjunction with its release on proprietary platforms, Mistral AI made the weights for the instruct-tuned version available for research and non-commercial use under the version identifier 24.11. The model architecture remains a dense transformer, continuing Mistral's approach of balancing high-capacity reasoning with operational efficiency. It supports a broad range of languages, including English, French, German, Spanish, Chinese, Japanese, and Arabic.