French AI startup Mistral has unveiled a new AI model designed for the Middle Eastern and South Asian regions. Named Mistral Saba, it marks the French company’s first attempt towards localised AI that has been tailored for specific cultures and linguistics. “One of the many custom-trained models to serve specific geographies, markets, and customers,” the official website described the new AI model. With AI spreading rapidly around the world, many users have expressed their desire for models that are not just fluent in regional languages but also get the regional parlance. “While larger, general-purpose models are often proficient in several languages, they lack linguistic nuances, cultural background, and in-depth regional knowledge required to serve use cases with strong regional context,” the company said on its official website. What is Mistral Saba? The new model from Mistral is a 24B parameter AI model trained on datasets that have been carefully curated from across the Middle East and South Asia. According to the company, Mistral Saba is capable of generating the most accurate and relevant responses, even better than models that are five times its size. The model is significantly faster and costs much less than other models in the same league. The company has said that the model can be used as a foundation to train highly specific regional adaptations. It is currently available as an API; however, it can be deployed locally within the security premises of its customers. The model is lightweight and can be deployed on single-GPU systems and responds at a speed of over 150 tokens per second. Mistral currently supports Arabic and some Indian-origin languages such as Tamil and Malayalam. According to the company, this ability endows the model with versatility in multinational use across these interconnect regions. When it comes to use cases, the company claims that Mistral Saba can be used to harness conversational support, domain-specific expertise, cultural content creation, etc.
