# Mistral Medium Frontier-class mid-range model line by [[Mistral AI]]. Initially released as Mistral Medium 3 in 2025 (proprietary, mid-range, cheap), updated to Mistral Medium 3.1 (August 2025), then re-positioned upward as the open-weights flagship merged model [[Mistral Medium 3.5]] in April 2026 — at which point Medium left the "cheap mid-range" slot entirely. ## Versions - Mistral Medium 3 (2025) — proprietary; 128k context; $0.4 / $2.0 per M tokens - Mistral Medium 3.1 (Aug 2025) — proprietary refresh - [[Mistral Medium 3.5]] (Apr 2026) — open weights (Modified MIT); 128B dense; 256k context; $1.5 / $7.5 per M tokens; flagship merged model behind Mistral Vibe and [[Le Chat]] ## Key characteristics - Proprietary; not open-weight - 128k context window - Multimodal: text and image inputs - API pricing: $0.4/M input tokens, $2/M output tokens - Performs at or above 90% of Claude Sonnet 3.7 across benchmarks at a fraction of the cost - Surpasses Llama 4 Maverick and Cohere Command A on benchmarks - Particularly strong in coding and STEM tasks ## Deployment - Minimum self-hosted: 4 GPUs - Hybrid, on-premises, or in-VPC deployment supported - Custom post-training and continuous pretraining available through Mistral's applied AI solutions ## Features - Chat completions, function calling, agents - Structured outputs, predicted outputs - OCR, document QnA, bounding box extraction - Fill-in-the-middle (FIM), embeddings - Audio transcription, moderation - Batch inference ## Availability - Mistral La Plateforme (AI Studio) - Amazon SageMaker - IBM WatsonX, NVIDIA NIM, Azure AI Foundry, Google Cloud Vertex ## References - https://mistral.ai/news/mistral-medium-3 - https://docs.mistral.ai/models/mistral-medium-3-1-25-08 ## Related - [[Mistral AI]] - [[Mistral Medium 3.5]] - [[Large Language Models (LLMs)]] - [[Mistral Small 4]] - [[Mistral OCR]]