The French startup Mistral AIwhich is backed by Microsoft, presented this Wednesday its new artificial intelligence (AI) models, which have been baptized as Ministerial 3B and Ministerial 8B. These models, the firm assures, «establish a new frontier in knowledge, common sense, reasoning, function calling and efficiency in the sub-10B categoryand can be used or adapted to a variety of uses.» Both models support a context length up to 128k and Ministral 8B has a «special interlaced sliding window attention pattern for faster and more memory-efficient inference.» «Our most innovative customers and partners have been increasingly requesting privacy-first local inference for critical applications like on-device translation, offline smart assistants, local analytics, and autonomous robotics. Ministraux were created to provide a low-latency, compute-efficient solution for these scenarios.«, details the company. Used in conjunction with larger language models, such as Mistral LargeMinistraux are also intermediaries for calling functions in multi-step agent workflows, and can additionally be tuned to handle input parsing, task routing, and API calls based on user intent in multiple contexts with «extremely low» latency and cost«.»In Mistral AI, We continue to promote the latest in cutting-edge models. It has only been a year since the launch of the Mistral 7B and yet our smallest model today (Ministral 3B) already surpasses it in most tests,» he concludes.