Amazon SageMaker JumpStart: 2026 year-to-date in review

This article reviews Amazon SageMaker JumpStart's 2026 progress, highlighting the onboarding of over 100 new foundation models from 30 providers and platform enhancements that streamline model discovery and deployment.

• Over 100 models onboarded from providers including Alibaba (Qwen), NVIDIA, Mistral, Google, IBM, and Cohere across reasoning, multimodal, creative AI, and specialized tasks
• Expanded NVIDIA portfolio with Cosmos3 world foundation models for video generation, physics simulation, and robotics applications
• Closed multimodal gap with vision-language, speech, and document understanding models from NVIDIA, Google, IBM, and Mistral
• Made creative AI accessible with image and video generation models from Black Forest Labs, Meta, and NVIDIA
• Prioritized efficient and quantized models like GLM-5.2 FP8, Phi-4-mini, and Granite 4.0 h-tiny for cost optimization
• Onboarded purpose-built models for autonomous workflows, OCR, retrieval, speech, and protein structure prediction
• Simplified Model Hub on SageMaker AI console with filtering by modality, publisher, size, and license
• Pre-configured endpoint deployment options for balanced, throughput, latency, and cost optimization
• Hugging Face deep link integration enables one-click deployment from Hugging Face to SageMaker Studio

SageMaker JumpStart continues reducing friction from model discovery to production deployment with expanded provider diversity and enhanced user experience.