Key Takeaways
1. Mistral AI has launched a new reasoning-based AI model called Magistral, which emphasizes thoughtful analysis over speed.
2. Magistral offers traceability in its reasoning process, supporting compliance in regulated sectors like finance, law, and healthcare.
3. The model serves as a French alternative to dominant US and Chinese AI systems, with support from President Emmanuel Macron for independent technology development.
4. Two versions of Magistral are available: Magistral Small (open-source, 24 billion parameters) and Magistral Medium (larger, subscription-based, available on cloud platforms).
5. Magistral operates in two modes: Flash Answers for quick responses and Think mode for detailed reasoning, allowing users to review the AI’s thought process.
Mistral AI, a company from France, has introduced its new AI model named Magistral, which is designed to solve problems through reasoning. This reasoning model takes longer to analyze issues, gather data, and formulate thoughtful responses compared to traditional non-reasoning AI systems. One key feature of Magistral is that it offers businesses traceability for its reasoning process, aiding in compliance with stringent laws and regulations, particularly in sectors like finance, law, and healthcare.
French Alternative in AI
Magistral serves as an option for organizations looking for alternatives to leading US AI models such as OpenAI, Perplexity, Anthropic, Meta, and Grok, as well as the Chinese model DeepSeek, which currently dominate the LMArena AI Leaderboard. Recently, French President Emmanuel Macron has endorsed Mistral AI as an illustration of independent technology that can help France maintain its autonomy in the developing AI landscape. He has also revealed plans for a national AI data center, a significant investment project estimated to cost between 30 and 50 billion Euros, aimed at bolstering France’s position in the AI domain.
Two Versions of Magistral
Mistral AI has launched Magistral in two distinct versions: the smaller open-source AI model known as Magistral Small, which has 24 billion parameters, and the larger Magistral Medium model available for businesses and individuals willing to subscribe or access via API. Additionally, the Magistral Medium model is set to be available on AI cloud platforms like Microsoft Azure AI.
Magistral Medium delivers performance comparable to the Chinese reasoning AI model DeepSeek R1, while the smaller Magistral Small is expected to perform at a lesser level. In Le Chat, Magistral primarily operates in its Flash Answers mode, which allows it to generate responses quickly. The company claims it can handle tokens of information up to ten times faster than rivals, although no concrete data has been provided. Nevertheless, many AI chatbot services typically return quick answers within a couple of seconds.
Modes of Operation
The Flash Answers mode lacks the detailed sources or references for its replies, so if users are looking for more thorough information, they are encouraged to use Think mode. This mode requires the AI to take more time in providing answers and clarifies each step of its reasoning, allowing users to scrutinize the AI’s thought process.
For those interested in researching and customizing the Magistral Small model from Hugging Face on their personal computers, it is advisable to have a high-performance graphics card, such as the Nvidia 5090 GPU available on Amazon.
Source:
Link
Leave a Reply