Overview

Mistral AI provides a cutting-edge platform for developing and deploying state-of-the-art generative AI models. It supports developers and businesses by offering customizable AI solutions that can be fine-tuned to meet specific needs.

Available Models

Premier Models

• Magistral Medium - Advanced model with 40k context
• Mistral Medium - Frontier-class multimodal model with 128k context
• Mistral Large - Top-tier model for high-complexity tasks
• Pixtral Large - Vision-capable large model with frontier capabilities
• Codestral - Cutting-edge language model for coding (256k context)
• Mistral Saba - Specialized model for Middle East and South Asia languages
• Ministral 8B/3B - Powerful edge models for on-device use cases

Free/Open Source Models

• Magistral Small - Small model with transparent multilingual capabilities
• Mistral Small - SOTA multimodal multilingual model with vision capabilities (Apache 2.0)
• Devstral Small - Best open-source model for coding agents
• Pixtral 12B - Vision-capable small model
• Mistral Nemo - Best multilingual open source model
• Codestral Mamba - First mamba 2 open source model

Research Models

• Mathstral - First math open source model (Apache 2.0)

Embedding Models

• Mistral Embed - State-of-the-art semantic embeddings for text
• Codestral Embed - Specialized embeddings for code representation

Moderation Models

• Mistral Moderation - Content moderation service for detecting harmful text across multiple policy dimensions

Services Not Supported in Dify

• Mistral OCR - OCR service for extracting text and images from documents (uses /v1/ocr endpoint, not compatible with Dify's model framework)

Model Versions

For maximum flexibility, we provide both Latest and Dated versions of each model:

🔄 Latest Versions (Recommended)

• Format: model-name-latest (e.g., mistral-large-latest)
• Behavior: Automatically points to the most recent version
• Use Case: Development, testing, and applications that benefit from automatic updates
• Advantage: Always get the latest improvements and features

📅 Dated Versions (Stable)

• Format: model-name-YYMM (e.g., mistral-large-2411)
• Behavior: Fixed to a specific model version
• Use Case: Production environments requiring reproducible results
• Advantage: Guaranteed consistency and predictable behavior

💡 Which Version to Choose?

• Use Latest for: Development, experimentation, staying current with improvements
• Use Dated for: Production deployments, research requiring reproducibility, compliance requirements

Magistral Models

Magistral models (Magistral Medium and Magistral Small) are designed for complex problem-solving and feature seamless integration with Dify.

🎯 Model Configuration

Magistral models are designed for step-by-step problem solving and analytical tasks.

🔧 Reasoning Mode Parameter

Magistral models support a reasoning_mode boolean parameter that controls how the model processes requests:

• reasoning_mode: true (Default): Enables reasoning mode with system prompt for step-by-step problem solving
• reasoning_mode: false: Disables reasoning mode for standard chat behavior without system prompt

This parameter is automatically transformed to the Mistral AI API prompt_mode parameter ("reasoning" or null) based on your selection.

Configure

After installation, you need to get API keys from MistralAI and setup in Settings -> Model Provider.