News
ReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior ReleaseReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior Release
EveryLLM
Mistral Saba

Mistral Saba

A 24B regional language model for Arabic and South Asian languages

Mistral Saba is a 24-billion-parameter language model built by Mistral AI for Arabic and Indian-origin languages, with particular strength in Tamil. Announced in 2025 as the company's first specialised regional model, it targets Middle Eastern and South Asian use cases and can run on a single GPU.

Share on WhatsApp

  • 24B parameters, trained on curated Middle East and South Asia data
  • Mistral AI's first specialised regional language model
  • Supports Arabic and Indian-origin languages, notably Tamil
  • Runs on a single GPU at over 150 tokens per second
  • Available by API and for local deployment inside customer premises

What it is

Mistral Saba is a language model from Mistral AI, sized at 24 billion parameters. It is the first in a line of specialised regional models from the company, aimed at Arabic and Indian-origin languages rather than general multilingual use. Mistral AI says it answers more accurately than models over five times its size while running faster and at lower cost. It is meant for work with strong regional and cultural context, such as conversational assistants, domain experts in energy, finance and healthcare after fine-tuning, and culturally grounded content such as educational material.

How it came about

Mistral AI introduced Mistral Saba as its first specialised regional language model, following the release of Mistral Small 3. The company says the model came out of close work with strategic regional customers on specific problems and bespoke use cases. Alongside regional models, it has begun training models for individual customers using their own private enterprise context, kept exclusive to those customers.

Inside the model

The model has 24 billion parameters and was trained on datasets curated from across the Middle East and South Asia. Mistral AI describes it as lightweight, in the same spirit as Mistral Small 3, able to run on a single-GPU system and to answer at speeds above 150 tokens per second. The company positions it as a base that customers can further train into highly specific regional adaptations, for example turning it into a specialist in energy, financial markets or healthcare within an Arabic language and cultural setting.

What it can do

Mistral AI claims the model gives more accurate and relevant answers than models over five times its size, while being faster and cheaper to run. It supports Arabic and many Indian-origin languages, and is described as especially strong in South Indian-origin languages such as Tamil. Reported uses include real-time Arabic virtual assistants, domain-specific experts after fine-tuning, and content that reflects local idioms and cultural references.

How to use it

The model is offered through an API, and Mistral AI also allows customers to deploy it inside their own security perimeter. It can be run locally on a single GPU. The company directs enterprise deployment enquiries to its sales team and points developers to the Mistral Saba API.

Why it matters

Mistral AI presents Mistral Saba as the start of a family of specialised regional language models, arguing that large general-purpose models miss linguistic nuance and cultural background. The company says the model is gaining traction among its customer base in the Middle East, with use cases emerging around assistants, domain experts and localised content.

Questions people ask about Mistral Saba

What is Mistral Saba for?

It is a regional language model aimed at Arabic and Indian-origin languages, built for use cases where cultural and linguistic nuance matters, such as assistants, domain experts and localised content.

How large is it and how fast does it run?

Mistral AI gives it 24 billion parameters. The company says it runs on a single GPU and responds at speeds above 150 tokens per second, and claims it beats models over five times its size on accuracy.

Which languages does it support?

Arabic and many Indian-origin languages, with particular strength in South Indian-origin languages such as Tamil, reflecting the cultural links between the Middle East and South Asia.

How can I use it?

Through the Mistral Saba API, or deployed locally within a customer's own security perimeter. Enterprise deployments are handled by contacting Mistral AI directly.

Can it be customised?

Yes. Mistral AI says it works as a base for training highly specific regional adaptations, for example a specialist in energy, financial markets or healthcare within an Arabic context.

Good reading and sources

    Where this page comes from

    There is no English Wikipedia article on Mistral Saba yet, so the facts on this page come from the sources listed above: mistral.ai/news/mistral-saba. Some are the subject's own pages, so its claims about itself are reported as its claims. EveryLLM wrote the text in its own words, with AI assistance, and compares it with the sources to make sure no passage is copied.

    It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.

    The weekly digest

    Every Monday: the model of the day, new pages on EveryLLM, and the conferences and anniversaries of the week ahead.