Mistral AI adds Medium 3 to its family of models, claiming low cost and high performance

Paris-based artificial intelligence startup Mistral AI today announced the release of a new model, Mistral Medium 3, which the company said outperforms competitors at significantly lower cost.

The company also introduced a new version of its AI assistant for enterprise customers, Le Chat Enterprise, powered by Medium 3, featuring enterprise search, agent building and tool connectors.

On performance, Mistral said the new model delivers performance surpassing models such as Meta Platform Inc.’s Llama 4 Maverick and enterprise models such as Cohere Inc.’s Command A. On price, it beats cost for leaders such as DeepSeek v3 with much lower cost through both application programming interface and self-deployed systems.

For enterprise use, the company said it delivers more than 90% or better benchmark performance of Anthropic PBC’s Claude 3.7 Sonnet. However, it can do it at one-eighth the cost or 40 cents per million input tokens and $20.80 per million output tokens, compared with Sonnet’s $3 per input token and $15 per output token.

Mistral said Medium 3 was designed as a frontier-class model and stands out in professional use cases, but works best for coding and science, technology, engineering and mathematics tasks, where it closes the gap with larger and much slower competitors.

Instead of having to choose between fine-tuning over API or self-deploying and customizing model behavior from scratch, Mistral said, Medium 3 offers a path to integrate intelligence comprehensively into enterprise systems.

Enterprise companies can continuously pretrain, fully fine-tune and blend the model into knowledge sources for domain-specific training and adapt it to both company culture and information systems. According to Mistral, beta customers across financial services, energy and healthcare have already harnessed the model to personalize business processes, workflows and analyze complex datasets.

The new model is available starting today via API on the company’s Mistral La Plateforme and Amazon SageMaker. It will also come soon to IBM watsonx, Nvidia NIM, Azure AI Foundry and Google Cloud Vertex.

Le Chat Enterprise AI assistant

Le Chat Enterprise builds on the company’s existing AI chatbot assistant with a focus on enterprise productivity tools and security into a unified AI platform.

It will provide enterprise teams a feature-rich platform that includes enterprise search, AI agent builders, custom data and tool connectors, document libraries, custom AI models and hybrid deployment capabilities.

Enterprise teams will be able to use Le Chat Enterprise to connect to all of their usual productivity apps such as Google Drive, Google Calendar, Gmail and Microsoft SharePoint and OneDrive in one place – with more connectors coming soon. Using this information, it can help organize external data sources, documents and web sources to provide relevant answers and provide a handy way to get up to speed in anything business-related.

With agent builders, workers can automate routine tasks by connecting to apps and business tools.

“Le Chat will enable your team to easily build custom assistants that match your own requirements — no code required,” Mistral said.

The company stressed that the enterprise-focused Le Chat is privacy-first and can be deployed anywhere, in public or private cloud, or as a service hosted on the Mistral cloud. All data connectors to enterprise tools remain fully protected and adhere to access controls.

Le Chat Enterprise is available today for enterprise teams in Google Cloud Marketplace, and will be coming soon to Azure AI and AWS Bedrock.

Image: SiliconANGLE/Mistral AI Le Chat