The French firm Mistral AI announced the launch of its first reasoning model, Magistral Small, published in the form of an open-source version and an enterprise version. The model generates responses through logical reasoning, integrates expertise in interdisciplinary fields and provides a transparent process of reasoning that can be traced and validated to compete with global AI leaders such as OpenAI and Google.
Magistral tailors the profession by supporting a verifiable reasoning process. The model addresses structured computing, process logic, decision tree and rule-based systems and is therefore applicable to business processes. According to Mistral, “the best thinking of humanity is not linear — it interwoven logic, insight, uncertainty and exploration. Logic language models enable us to enhance complex thinking and deep understanding and entrust them to artificial intelligence.”
Magistral Small, an open source model with 24 billion parameters, supports the autonomous deployment of users using the Apache 2.0 licence agreement, and is now on board Hugo Face. The model also optimizes multilingual reasoning, particularly in English, French, Spanish, German, Italian, Arabic, Russian and Chinese, and is suitable for development communities and academic research.
Magistral Medium, an enterprise-level closed-source model with unpublished parameters, introduced a reflection mode and Flash Answers in Le Chat to support Amazon SageMaker, which will be extended to IBM WatsonX, Microsoft Azure AI and Google Cloud Marketplace in the future. The model scored 73.6 per cent (70.7 per cent compared to DeepSeek R1) in the AME 2024 mathematical reasoning benchmark test, which was 10 times faster than the partial competition.
The central strength of Magistral lies in the transparent chain of reasoning (Chain-of-Thought, COT), which can demonstrate complete logical steps, including confidence scoring and error amendment processes, in areas that require high compliance, such as law, finance, health care and government. For example, when dealing with complex legal cases or financial modelling, Magistral can provide a path of retroactivity to ensure that conclusions meet regulatory requirements. In addition, its multilingual support and interdisciplinary knowledge integration (e.g. programming, mathematics, science) makes it uniquely competitive in European markets.
Mistral AI was founded in 2023, valued at $6 billion and supported by French President Marlon and Microsoft, with its open-source strategy to make it the European AI flag. Magistral’s release is a direct response to reasoning models such as OpenAI O3, Gemini 2.5 Pro of Google and DeepSeek R1 of China. According to the co-founder of Mistral AI, Guillaume Lample, “Magistral, through a transparent reasoning process, enables users to understand the decision logic of AI and to meet compliance needs”.
On 11 June 2025, at the GTC Paris Summit, the CEO Wong In-hoon announced cooperation with Mistral AI to create Europe ‘ s largest AI cloud, based on the DGX Cloud Lepton platform and 18,000 Grace Blackwell chips, and is expected to go online in early 2026. This cooperation will enhance Magistral ‘ s computing support and promote its application in the industrial and scientific fields.
