In recent months, we have seen many artificial intelligence models focused on working on laptops and mobile devices, including MobileLLM and SmoILM. Finally, the France-based AI company Mistral has joined this trend with its models called les Ministraux.

Les Ministraux includes the Ministral 3B and Ministral 8B models. It is worth noting that both models have a context window of 128,000 tokens. This means the models can take inputs approximately as long as a 50-page book. According to the company, Mistral’s clients needed a localized, privacy-focused approach for applications such as on-device translation, offline smart assistants, local analytics, and autonomous robotics. Les Ministraux was developed to provide a computationally efficient and low-latency solution for these scenarios.

According to graphs shared by Mistral, the Ministral 3B and Ministral 8B outperformed similar Llama and Gemma models, as well as Mistral’s own 7B model, across various AI benchmarks. The benchmarks are designed to evaluate abilities such as following instructions and solving problems.

With only 3 billion parameters, the Ministral 3B performs better than Mistral’s original 7 billion parameter model in most comparisons.

The Ministral 8B, with its performance, can compete with larger models.

The compact models have diversified the cloud approach we have seen in the market so far. Running AI models locally on devices ensures that sensitive data remains in the user’s possession. If you’re considering running AI models on your own computer, it is beneficial to check out our article on tools that allow you to run AI models on your computer.

As of today, the Ministral 8B is available for download only for research purposes. For commercial licensing of the Ministral 8B or Ministral 3B, you need to contact Mistral. Developers can use the Ministral 3B and Ministral 8B through Mistral’s cloud platform, La Platforme, and other cloud services the initiative will partner with in the coming weeks. The Ministral 8B is billed at 10 cents per million output/input tokens, which is equivalent to approximately 750,000 words. The Ministral 3B costs 4 cents per million output/input tokens.


Source