Mistral Nemo is a 12B model developed in collaboration with Nvidia, offering outstanding reasoning and coding performance, easy to integrate and replace.

Mistral Small is a cost-effective, fast, and reliable option suitable for use cases such as translation, summarization, and sentiment analysis.

Mistral Large is the flagship model, excelling in multilingual tasks, complex reasoning, and code generation, making it an ideal choice for high-end applications.

Codestral is a cutting-edge generative model focused on code generation, optimized for intermediate filling and code completion tasks.

Pixtral Large is an open-source multimodal model with 124 billion parameters, built on Mistral Large 2. This is the second model in our multimodal family, showcasing cutting-edge image understanding capabilities.

The Pixtral model demonstrates strong capabilities in tasks such as chart and image understanding, document question answering, multimodal reasoning, and instruction following. It can ingest images at natural resolutions and aspect ratios and handle an arbitrary number of images within a long context window of up to 128K tokens.

Ministral 3B is Mistral's top-tier edge model.

Ministral 8B is Mistral's cost-effective edge model.

Mistral 7B is a compact yet high-performance model, excelling in batch processing and simple tasks such as classification and text generation, with good reasoning capabilities.

Mixtral 8x7B is a sparse expert model that leverages multiple parameters to enhance reasoning speed, suitable for handling multilingual and code generation tasks.

Mixtral 8x22B is a larger expert model focused on complex tasks, providing excellent reasoning capabilities and higher throughput.

Codestral Mamba is a language model focused on code generation, providing strong support for advanced coding and reasoning tasks.