Mistral Ministral-3 3B and 8B Instruct 2512 land on AWS SageMaker JumpStart

On September 14, 2026, Mistral AI's Ministral-3-3B-Instruct-2512 and Ministral-3-8B-Instruct-2512 models became available on Amazon SageMaker JumpStart, expanding the catalog of foundation models available to AWS customers. Ministral-3-3B-Instruct-2512 is engineered for ultra-lightweight edge deployment, pairing a 3.4-billion-parameter language model with a 0.4-billion-parameter vision encoder; the model fits in 8GB of VRAM at FP8 precision and supports a 256K-token context window. Ministral-3-8B-Instruct-2512 shares the same vision, multilingual, agentic, and function-calling capabilities as the 3B sibling but scales the language model to 8.4 billion parameters, fitting in 12GB of VRAM at FP8 with an interleaved sliding-window attention pattern for faster, more memory-efficient inference. Both ship under Apache 2.0 with vision analysis, multilingual instruction following (English, French, Spanish, German, Chinese, Japanese, Korean, Arabic, and more), strong system-prompt adherence, and native function calling with structured JSON output. Mistral's flagship Mistral Large 3 (675B total / 41B active parameters) launched December 2025 at $0.50 per million input tokens and $1.50 per million output tokens, about a quarter of the equivalent Mistral Large 2 pricing. On the same day, NVIDIA's Qwen3.6-35B-A3B-NVFP4 and Alibaba's Wan2.1-T2V-1.3B-Diffusers also landed on SageMaker JumpStart, giving AWS customers one-click deployment access to an NVFP4-quantized Qwen3.6-35B-A3B for agentic coding and long-context reasoning as well as a consumer-grade text-to-video model. The simultaneous arrival of Mistral's Ministral-3 family and Alibaba's Qwen3.6-35B-A3B-NVFP4 on SageMaker JumpStart is being read as a coordinated move by Mistral and Alibaba Cloud to expand developer reach through the AWS ecosystem, picking up the pace of Mistral's commercial rollout following the September 8 announcement of its €3 billion Series D.