Updates·December 3, 2025, 22:38
Mistral Medium 3: cheap multimodal model for businesses
AI-generated and checked against the sources listed below.
Mistral Medium 3 understands text and images, costs $0.40 per million input tokens and is aimed at businesses.

What is new
French AI company Mistral has launched Mistral Medium 3, a multimodal model. That means it can understand both text and images. The model is built for businesses and is meant as a cheaper alternative to the large models from OpenAI, Google and Anthropic.
At the same time, Mistral launched Le Chat Enterprise, a chatbot for companies. It has a builder for AI agents (small assistants that carry out tasks on their own) and can connect to Gmail, Google Drive and SharePoint. Support for Anthropic's MCP standard, which connects AI assistants to a company's data, was to come later.
The model was available via Amazon SageMaker from May 7, 2025. Microsoft Azure AI Foundry and Google Vertex AI were next in line. The sources do not say whether there are special terms for Denmark or the EU.
What is clever
Companies can run the model wherever they want: in any cloud or on their own servers with four or more graphics cards (GPUs). That is interesting for companies that do not want to send sensitive data to an American service.
The model can also be further trained on the company's own data. Then it answers more precisely about what applies there specifically. Mistral especially highlights coding, scientific tasks and understanding images and documents. Beta testers in banking, energy and healthcare have tried it for customer service, workflow automation and data analysis.
Cheaper or better?
On price, there is little to fault. Via Mistral's API, Medium 3 costs $0.40 per million tokens (pieces of text) in and $2 per million tokens out. The monthly price in kroner for ordinary users is not stated in the sources.
Mistral says the model handles "at least 90 percent" of what Anthropic's more expensive Claude Sonnet 3.7 can do across common tests (benchmarks). According to Mistral, it also beats Meta's Llama 4 Maverick and Cohere's Command A. Mistral's own tests also put it ahead of GPT-4o and Claude 3.7 Sonnet in several areas. Remember that the figures come from the company itself. We have not seen independent measurements here.
So it should not be seen as the world's strongest model, but as a model that offers a lot for the money.
What it is good at
Medium 3 suits companies that want to use AI at scale without paying top price. That includes customer service, analysis of documents and images, programming help and automation of routine tasks.
It is also interesting for companies with data control requirements, because it can run on their own servers. Ordinary users get closest to it through Mistral's chatbot Le Chat.
Note: The Traders Union source the story is based on does not mention the model's name. The details here come from Mistral's launch of Medium 3 as covered by TechCrunch.
Sources
Get the week's AI news in your inbox
Choose your level, topics and length. One email a week, unsubscribe at any time.
Subscribe to Promptly Newsletter



