Join our daily and weekly newsletters for the latest updates and the exclusive content on AI coverage. Learn more
The startup of French AI Mistral has collected private funding cargoes, but has not yet Fall the graphics of use of the upper AI overall, in particular Regarding the adoption of companies and developers.
But this can change from today: the The company has just unveiled the corporate catAn AI Unified Assistant platform designed for productivity and confidentiality on a business scale, Propelled by its new average model 3 This surpasses the largest to a cost fraction (here, “larger” refers to the number of parameters or to internal model parameters, which generally denote more complexity and more powerful capacities, but also take more calcular resources such as GPUs to operate).
The Enterprise cat is an assistant and a chatgpt type competitor built from zero for data protection, audit and cross -application support
Available on the web and via mobile applications, the Enterprise cat is like a competitor Chatgpt, but that is specifically built for businesses and their employees, taking into account the fact that they will likely work through A series of different applications and sources of data.
It is designed to consolidate AI’s features in a unique environment and focused on confidentiality that allows deep personalization, interfunctional workflows and rapid deployment.
Among its main characteristics that will interest business owners and technical decision -makers are:
- Business search in private data sources (Google Drive of your business, SharePoint, Gmail, etc., without exposing or publishing information to third parties)
- Documents of documents with sub-elevation and quotation capacities
- Personalized connectors and agent manufacturers for the automation of tasks without code
- Personalized and personalization -based customization model integrations
- Hybrid deployment options with public cloud support, private VPCs and on -site accommodation
The Enterprise cat supports transparent integration into existing tools and workflows. Companies can build AI agents adapted to their operations and maintain full sovereignty over deployment and data, shamelessly suppliers.
The platform confidentiality architecture complies with strict access controls and supports complete audit logging, guaranteeing data governance for regulated industries. Companies also take total control over the AI battery – infrastructure and platform features at personalization and user interfaces at the model level.
And taking into account the general suspicions of certain Western societies and governments through China and its increasing library of powerful open source models of companies there, associated with the location of Mistral in the European Union and with close laws of data protection that it must follow (General regulations for the protection of alias RGPD Data and the EU ACT AI), The new Mistral offer Le Chat Enterprise could be attractive for many companies with more strict security and data storage policies (in particular medium and inherited companies).
Mistral also deploys improvements in his chat pro and team plans, targeting individuals and small teams looking for productivity tools supported by his linguistic models. All levels benefit from the basic capacities introduced into the Enterprise cat.
Mistral Medium 3 surpasses GPT-4O and even Claude 3.7 Sonnet on key references and is available via API and on site
Mistral Medium 3 introduces a new level of performance in the company’s range of models, positioned between light and large -scale models. This is a proprietary model, which means that unlike the previous mistral versions, it is not available under an open source license and must be used via the website and the API of Mistral or that of its partners.
Designed for the use of the company, the model offers more than 90% of Claude 3.7 Sonnet’s reference performance, but at an eighth cost – $ 0.40 per million entry tokens and $ 20.80 per million exit tokens, compared to $ 3 / $ 15 of Sonnet for entries / exits.
References show that Mistral Medium 3 is particularly strong in software development tasks. In coding tests like Humaneval and Multipl-e, it corresponds or exceeds the Claude 3.7 Sonnet and GPT-4O of Openai models.
According to third-party human assessments, he surpasses Llama 4 Maverick in 82% of coding scenarios and exceeds the Command-A in almost 70% of cases.
The model also works competitively between languages and methods. Compared to Llama 4 Maverick, he has higher victory rates in English (67%), French (71%), Spanish (73%) and Arabic (65%) and multimodal performance with notable scores in tasks like Docvqa (0.953), AI2D (0.937) and Chartqa (0.826).
Mistral Medium 3 is optimized for business integration. It supports hybrid and on-site deployment, offers personalized post-training and easily connects to commercial systems. According to Mistral, it is already used in beta version by organizations in sectors such as financial services, energy and health care to feed workflows specific to the field and customer -oriented solutions.
Mistral Medium 3 is now accessible via The API of the Mistral plate And Amazon SagemakerWith a support that soon arrives on IBM Watsonx, Nvidia Nim, Azure Ai Foundry and Google Cloud Vertex.
Meanwhile, the Enterprise cat is available in the Google Cloud MarketplaceAnd will start a little on Azure AI and Aws Boudrock.
For those who are ready to explore the assistant experience, the cat is available to cat.mistral.aiAs well as in the App Store and Google Play Store, without credit card required to start.
By combining a high-efficiency model with a customizable corporate platform, Mistral AI made a concerted push to reduce barriers to the adoption of scalability of the confidentiality of confidentiality in the business world.
