The Mistral (2023) family consists of open-weight transformer-based large language models beginning with Mistral 7B, released by Mistral AI in September 2023. The models feature architectural innovations including Grouped-Query Attention and Sliding-Window Attention for efficient processing of long contexts up to 8,192 tokens. Various derivatives have been fine-tuned for specific applications including dialogue, instruction following, and code generation using techniques like Direct Preference Optimization and progressive learning from explanation traces.