Mistral has launched its Mistral 3 series, featuring a frontier model and efficient, smaller models tailored for offline enterprise applications and customisation. The company aims to demonstrate that smaller, fine-tuned AI models can outperform larger, closed-source alternatives. The Mistral Large 3 is designed to operate in multiple languages. The company also released nine 'Ministral' models that can be run directly on devices without internet connection.
The Mistral Medium 3 is optimised for enterprise use, delivering performance at a lower cost. It excels in coding, STEM reasoning and multimodal understanding, supporting hybrid and self-hosted deployment. The model is designed for software development, data analysis and reasoning.
Amazon Bedrock has added 18 fully managed open weight models, including the Mistral Large 3 and Ministral 3 models. The Mistral Large 3 is optimised for long-context, multimodal and instruction reliability. The Ministral 3 is edge-optimised for single GPU deployment with language and vision capabilities.
