Mistral released its first Reasoning-Centenary Artificial Intelligence (AI) model on Tuesday. Dubbed magistral, large language model (LLM) is available in two variants – small and medium. While magistral small is an open-source model, the middle version is an enterprise-centered closed model. Both models are capable of multi-step region and show transparent chain-off-three (cot). The French AI Startup highlighted that the magistral is designed to take research, strategic plan and data-driven decisions. A preview version of the Reasoning model can also be tried through the company’s Le Chat platform.
Mistral’s Magistrate Small is available with Apache 2.0 license
In a news room PostThe Paris -based AI firm announced the release of two magistral models. Magistral Small, Open Wet LLM, Mistral is available for download and deployment through Hugging Face Entry Apache 2.0 with license. The model can be used for educational and commercial use cases.
On the other hand, Magistral Medium is a proprietary model that can be accessed through Amazon Sammekar. It will soon be hosted at IBM Watsonx, Azure AI and Google Cloud Marketplace. A preview version of the model can also be experienced in Le Chat or through API on the LA platform.
Coming for the model details, the magistral is a 24 billion parameter model. The parameters of the enterprise version have not been revealed by the company.
Focusing on the capabilities of the magistral medium, the model has been claimed to score a 73.6 percent on AIME2024 compared to the Dipsek-R1. On the other hand, the small version is said to have scored 70.7 percent, Mistral Said.
Both models can argue basically, and COT works in many global languages. The company said that magistral can argue in English, French, Spanish, German, Italian, Arabic, Russian and simplified Chinese languages.
With most logic models, magistral structured calculations, programmatic logic, as well as decisions can handle trees and rules-based systems.
Mistral says that enterprises and users such as Finance, Healthcare, Government and Law are working in important areas, will achieve tracedable arguments within the model, allowing them to investigate the logical steps taken to reach a response. This will allow users to audit the reactions that are more sensitive than others, the company said.