Mistral Large 2 (24.07) is a flagship dense large language model released by Mistral AI in July 2024. Featuring 123 billion parameters, the model is designed to provide high-performance reasoning and coding capabilities while remaining efficient enough for single-node inference. It represents a significant evolution of the Mistral Large family, focusing on cost-efficiency and performance benchmarks competitive with major frontier models.
The model is highly proficient in programming, having been trained on over 80 coding languages including Python, Java, C++, and JavaScript. It also demonstrates advanced mathematical reasoning and logical problem-solving abilities. Mistral AI specifically optimized the model to reduce hallucinations and ensure more concise, accurate responses by training it to acknowledge when it lacks sufficient information to answer a prompt.
Mistral Large 2 supports a context window of 128,000 tokens and offers extensive multilingual support for dozens of languages such as French, German, Spanish, Arabic, Hindi, Chinese, and Japanese. It is further equipped with native function calling and structured JSON output capabilities, making it suitable for complex agentic workflows and integration with external software tools.