Mistral Medium 3

Overview

Mistral Medium 3 is Mistral AI's mid-tier model, released May 7, 2025. It sits between the cost-efficient Mistral Small 3-series and the frontier-level Mistral Large 3, offering a balanced option for applications that need more capability than the smallest models without the full cost of the flagship. At $0.40/$2.00 per million tokens with a 131K context window, it is priced competitively against other mid-tier options.

Specifications

• Developer: Mistral AI
• Model String: mistral-medium-latest (check Mistral docs for versioned string)
• Release Date: May 7, 2025
• Type: Large Language Model, Open-Weight
• Context Window: 131,072 tokens (131K)
• License: Apache 2.0
• Access: Mistral API (la Plateforme), Azure AI Foundry, AWS Bedrock, self-hosted
• Pricing: $0.40 per million input tokens / $2.00 per million output tokens

Capabilities

Balanced Performance: Designed for use cases that need more than a small model can offer but don't require frontier-level reasoning. Solid performance on instruction following, text generation, summarization, and general Q&A.

Open-Weight Flexibility: Apache 2.0 licensed — can be fine-tuned and self-hosted freely without usage restrictions.

Multilingual: Handles multiple European and major world languages effectively, consistent with Mistral's multilingual focus.

Limitations

With 131K context, it is behind the 256K of Mistral Large 3 and the 1M+ context of models from Anthropic, OpenAI, and Google in the same price range. For tasks needing deep reasoning or complex multi-step analysis, Mistral Large 3 is the stronger choice.

Recent Developments

• May 7, 2025 Launch: Part of Mistral's 2025 portfolio expansion alongside Mistral Small 3.1 and other models.
• Superseded in some use cases by Mistral Large 3: With the December 2025 launch of Mistral Large 3 at $0.50/$1.50, the price gap between Medium 3 and Large 3 narrowed significantly, making Large 3 the preferred choice for most use cases.

Last Updated

February 26, 2026