Enterprise leaders have joined a reliable program for nearly two decades. VB transform brings people to develop real enterprise AI strategies. Get more information
European AI Power House MrA new family of large language models (LLM) that makes the company the first sign to enter the growing competitive space of “reasoning”, or models that take time to consider their thinking to catch errors and solve their thinking more complex tasks than the basic text -based LLM.
The announcement offers a strategic dual release: a powerful, proprietary magistrate medium for enterprise clients, and, in particular, the parameter of 24 billion open source version, magistrate smaller.
The latter release appears to be calculated to strengthen the company’s affiliation with its basic roots, after a period where it faced criticism of the tilt in Medium 3 for more closed, proprietary models such as the proprietary model, which began in May 2025.
Return to the open source roots
The move will undoubtedly be celebrated by developers and the wider AI community, Mysterol Magistral Small is releasing the permit under Open Source Apache 2.0 License.
This is an important detail. Unlike more binding licenses, Apache 2.0 allows anyone to use, edit and distribute the model’s source code freely for commercial purposes.
This has empowered the construction and deployment of Mr. Licensing fee or vendor Lock, a similar licensing fee or vendor Lock, without their fears.
In view of the context, this open view is especially important. Although Mr. Master made his reputation on powerful open models, some of the current Medium 3’s purely proprietary offer raised concerns from some circles of the open source community, fearing that the company is moving towards more closed environmental systems, like rivals like Open AI.
Under such a legitimate license, the release of the magistrate Small acts as a powerful anti -rhetoric, which confirms the dedication to armed the open community with modern tools.
Competitive performance against strong enemies
The misunderstanding is not just talking about a big game. It came with receipts. The company released a benchmark that presented its predecessor, Mr. Medium 3, and Deep Sek to a Magistral Medium against rivals. The results show a model that is extremely competitive in the field of reasoning.
On the AIME-24 mathematical benchmark, the Magistral Medium has scored 73.6 % impressive on the neck and neck with its predecessor and significantly forwarded the DiPsic models. When the majority use voting (a technique where the model produces multiple answers and is the most common choice), its performance on Aime-24 jumps at 90 %.
The new model also has other demand tests, including GPQ Diamond, a graduate -level question -answering benchmark, and the LI Li Leo Code bench of coding challenges.
Although DiPsic-V3 performs strongly at some of the standards, the Magistral Medium permanently proves itself an advanced reasoning model, which verifies the claims of Mr.
Enterprise Power
While the Magistral Small meets the world of open source, but benchmark confirmed Magistral Medium The enterprise is square.
It is capable of Mr. Li Chat Interface and LA Plays API, it provides advanced performance required for important tasks of the mission.
Mr. is making this model available on major cloud platforms, including the Amazon Sage Maker, which includes Azure AI, IBM Watsonics, and the Google Cloud Market Place.
This dual release strategy allows the misunderstanding to keep your cake and eat it too: promoting its most powerful, performance testing technology for corporate clients, promoting dynamic ecosystem around their open models.
The cost of the cost
When it costs, Mysterol is also positioned as a separate, premium offer, compared to its models.
Per 2 million input token and 5 million in 5 million output token, it represents a significant price increase from the old wrong medium medium 3, which costs only 40 0.40 for input and $ 2 for output.
However, when it is placed against its external rivals, the strategy of determining the pricing of the magistrate medium is very aggressive. Its input cost is similar to the latest open model of the open and Gemini sits in the 2.5 Pro range, yet its $ 5 output price reduces both, which costs more than $ 8 and $ 10 respectively.
Although this is much more expensive than special models such as DiPsic-controversy, this is a great price for Anthropic’s flagship cloud Ops 4, making it a great price for consumers to achieve the latest reasoning without paying the market for absolute prices.
The reasoning you can see, understand and use
The three basic benefits with Mr. Mulla Magistral Line are forwarding: transparency, multi -linguistic and speed.
The “black box” of many AI models is designed to produce a “chain -off thinking” of a trap, a tragic one. This facilitates users to follow the logical path of the model, which is an important feature for high -steaks professional fields such as law, finance, and health care, where results should be verified.
In addition, these arguments are global. Mr. has emphasized the model’s “multi -linguistic skills”, highlighting high sincere performance in languages, including French, Spanish, German, Italian, Arabic, Russian and easy Chinese.
On the performance front, the company claims to promote a significant pace. The feature of a new “Think Mode” and “Flash Answers” in Le Chat allegedly enables the Magistral Medium to get up to 10 times in the token throptt of competitors, which facilitates real -time reasoning on the previous scale.
From the code general to the creative strategy and beyond
Applications for the magistrate are wide. From financial modeling and legal analysis to software architecture and data engineering, Mistress is targeting any use of any use that demands precision and structure. Even the company exhibited the ability to produce a shot physics simulation, showing the grip of complex systems.
But this is not the whole business. Mr. also recommends the model as a “creative partner” to tell the story and tell the story, which is capable of creating work that is either extremely integrated or, as the company has said, “Happily lifted.”
With the magistrate, Mr. AI is not only making a strategic game to compete, but also guides the AI’s next border. When re -engulfing its open source base with a powerful, legitimately licensed model, when placing the envelope on the performance of the enterprise grade simultaneously, the company is indicating that the future of AI will be powerful in the future and, in a meaningful way.
