Ministral-3-3B-Instruct-2512 and Ministral-3-8B-Instruct-2512 models now available on Amazon SageMaker JumpStart

This article announces the availability of Mistral AI's Ministral-3-3B-Instruct-2512 and Ministral-3-8B-Instruct-2512 models on Amazon SageMaker JumpStart.

• Ministral-3-3B: 3.4B language model with 0.4B vision encoder, fits in 8GB VRAM (FP8), supports 256K-token context window
• Ministral-3-8B: 8.4B language model with 0.4B vision encoder, fits in 12GB VRAM (FP8), offers frontier-class capabilities
• Both models support multimodal understanding, multilingual instruction following, function calling, and structured JSON output
• Designed for edge deployment and resource-constrained environments with high performance and scalability
• Deploy via SageMaker JumpStart console or Python SDK with just a few clicks

These compact, vision-capable models expand AWS customers' foundation model options for efficient AI deployment on edge and resource-constrained infrastructure.