Mixtral 8x22B Instruct

MistralMistralFrance

A large-scale Mixture-of-Experts (MoE) model composed of 8 experts, each with 22B parameters, enabling efficient inference through sparse activation. It excels at instruction following, dialogue, and complex reasoning tasks while maintaining a relatively low inference cost.