Artificial Intelligence

Amazon SageMaker JumpStart Expands With 100 Models in 2026

Amazon SageMaker JumpStart has significantly expanded its artificial intelligence catalog by onboarding over 100 new models from 30 distinct providers during the first three quarters of 2026.

Radwan AbdulhadiEditor and technology writer

Published: October 20, 2026 · Updated: October 20, 2026 · 3 min read

Amazon SageMaker JumpStart Expands With 100 Models in 2026 — Image: AWS Machine Learning Blog

Catalog Expansion and Provider Highlights

Amazon SageMaker JumpStart serves as a machine learning hub within Amazon SageMaker AI, offering access to a curated catalog of AI models, solution templates, and example notebooks. Between January and September 2026, the hub onboarded over 100 new models across 30 distinct providers to address accelerating open source releases and growing customer demand for choice. More details are available on the official AWS Machine Learning Blog.

The expansion features a diverse set of contributors across the artificial intelligence landscape. Alibaba's Qwen family contributed 29 models spanning reasoning, mixture-of-experts, and agentic workflows. NVIDIA added 17 models, including the Cosmos3 world foundation model family and specific offerings like the Nemotron 3.5 Lightning.

Multimodal and Physical AI Integration

Customer workloads increasingly require models that natively process multiple data types beyond text. In 2026, SageMaker JumpStart prioritized onboarding models capable of handling vision, speech, and document understanding. NVIDIA introduced the Nemotron 3 Nano Omni for text, vision, and speech processing, alongside the Nemotron 3.5 Super VL for vision-language tasks.

In addition to conversational and vision models, NVIDIA's Cosmos3 world foundation models brought video generation, physics simulation, and action planning into a unified architecture suited for physical AI applications such as robotics and digital twins. Other multimodal and creative additions included Meta (Muse Spark) and Reflection (Beam) integration.

Compact Models and Purpose-Built Tasks

To help organizations optimize inference costs and reduce latency in production environments, JumpStart onboarded various compact and quantized models. Offerings such as Zhipu AI GLM-5.2 FP8, JetBrains Mellum2-12B, and OpenBMB MiniCPM5-1B allow developers to execute targeted workloads efficiently on standard instance types without relying exclusively on frontier-scale architectures.

Furthermore, the hub expanded its catalog of purpose-built models optimized for specific tasks rather than generalized usage. Newly integrated solutions include specialized OCR tools for document text extraction, reranking models for retrieval-augmented generation pipelines, and OpenFold3 for protein structure prediction in scientific computing.

Platform Enhancements and Accessibility

Alongside model onboarding, the platform introduced several structural enhancements to streamline deployment workflows. Users can now utilize a simplified Model Hub, pre-configured deployment options tailored for cost, latency, or throughput, and a Hugging Face deep link integration that connects directly to external model registries.

These updates allow engineering teams to browse foundation models complete with architecture details, benchmarks, and licensing information, before deploying them with one click or fine-tuning them via low-rank adaptation and other serverless customization methods.