Agentic workflows and API tools
Mistral AI released its Agents API to support building agentic AI applications. This cloud-based service provides persistent memory to maintain context across conversations and enables agentic orchestration for workflows like agent hand-offs or sequential execution. The API includes built-in tools for code execution, web search, image generation, and document RAG. It also incorporates Model Context Protocol support for integrating external data sources and tools. Mistral’s API provides the best way for organizations to deploy agentic workflows specifically using Mistral foundation models. Developers use the Mistral AI library to access these capabilities. One limitation remains that agents often experience flakiness or require long periods of human-in-the-loop interaction during deployment. Mistral provides a cookbook to demonstrate these functionalities, including a GitHub code writing agent using the Devstral model and a financial agent for stock price analysis. A multi-agent system for earnings call analysis integrates Mistral OCR for document processing. This API complements the existing Chat Completion API by providing a dedicated framework for agent use cases. The API allows developers to add and remove agents dynamically to create complex workflows.
Microsoft infrastructure and sovereignty
Microsoft agreed to spend billions of dollars on Mistral’s computing infrastructure in Europe. This deal allows Microsoft Azure customers to develop software using Mistral’s data centers in France. By providing access to Mistral’s models on Azure Local, Microsoft provides businesses in regulated industries an alternative to US-controlled infrastructure. This partnership helps Mistral meet its goal of 1 gigawatt of compute capacity by 2030. Microsoft and Mistral are currently working on a joint go-to-market plan. The partnership addresses European concerns about AI sovereignty and control over where systems and data reside. Microsoft President Brad Smith stated that by putting Mistral’s models on Azure Local and on Mistral’s computational capacity, the companies combine American and European technology in a way that provides continuous and assured access for all European enterprise customers during this expansion. Microsoft’s partnership does not include any new equity investment in Mistral. The agreement aims to help businesses that require local data processing to avoid relying on US-controlled systems. This agreement also provides more capacity in Europe for Microsoft.
Model performance and pricing
Mistral’s API platform provides access to various model families. Mistral Large 2 performs well in coding tasks, surpassing Gemini 1.5 Pro on HumanEval benchmarks. Mistral Large 3 provides a 262K token context window and manages complex reasoning tasks. These models remain available via API only. Mistral Large 3 has an input price of $0.50 per million tokens and an output price of $1.50 per million tokens. You might find the pricing structure more predictable than US-based competitors because Mistral emphasizes transparent costs. While Mistral Large 2 performs well in coding, it has a smaller 128K context window compared to the 2 million token capacity of Google’s Gemini 1.5 Pro. Mistral also provides Ministral 14B and Ministral 3B for different needs. The Ministral 14B model has an input price of $0.20 per million tokens and an output price of $0.20 per million tokens. The Ministral 3B model has an input price of $0.10 per million tokens and an output price of $0.10 per million tokens. Codestral is another option with an input price of $0.30 per million tokens and an output price of $0.90 per million tokens. Mistral Large 2 has an input price of $0.70 per million tokens and an output price of $2.10 per million tokens. Mistral Large 3 uses a context window of 262K tokens. The Ministral 14B model has a 262K token context window. The Ministral 3B model has a 131K token context window. Codestral has a 256K token context window. Mistral Large 2 has a 128K token context window.
| Model | Input Price | Output Price | Context Window |
|---|---|---|---|
| Mistral Large 2 | $0.70 | $2.10 | 128K |
| Mistral Large 3 | $0.50 | $1.50 | 262K |
| Ministral 14B | $0.20 | $0.20 | 262K |
| Ministral 3B | $0.10 | $0.10 | 131K |
| Codestral | $0.30 | $0.90 | 256K |
Will Mistral maintain its competitive pricing as more American providers expand their European infrastructure?
