Three new arrivals in SageMaker JumpStart! GLM-5.2 FP8, Nemotron-Nano-12B-v2, and GLM-OCR are now available!

Hi, I'm Shii-chan! Today I get to share some exciting news about new models landing in AWS's SageMaker JumpStart, so let's dive right in!

AWS What's New aws.amazon.com

What was announced?

AWS's What's New page announced that three new foundation models are now available on Amazon SageMaker JumpStart: Z.ai's GLM-5.2 FP8, Z.ai's GLM-OCR, and NVIDIA's Nemotron-Nano-12B-v2! Each model brings a different strength to the table, covering long-horizon agentic engineering, efficient hybrid-architecture reasoning, and advanced document understanding.

Why it matters

SageMaker JumpStart already hosts a long list of foundation models, but what makes this addition stand out is that each of the three new models is specialized for a different job. Now, within a single platform, you can pick a long-context coding agent model, a resource-efficient reasoning model, or a document-digitization model depending on what you need.

What changes

• Engineers who want to hand off a full software development workflow — from requirements to deployment — as a single long-running task
• Enterprise teams who want to optimize inference cost and speed without giving up accuracy
• Anyone who needs to digitize and structure scanned documents, handwritten notes, or academic papers full of equations

With this update, all of these people can deploy the model that fits their use case in just a few clicks from the SageMaker model catalog, or programmatically through the SageMaker Python SDK — that's the biggest change here!

Dive Deep

Let's look at each model in more detail.

GLM-5.2 FP8 (Z.ai)

• Optimized for long-horizon agentic engineering tasks
• A major leap over its predecessor, GLM-5.1, and the first version with a genuinely usable 1M-token context window
• Keeps project-level engineering context in view across long-running tasks, follows engineering standards consistently, and can complete a full development workflow — from requirements to deployment — in a single task

Nemotron-Nano-12B-v2 (NVIDIA)

• Uses a hybrid Mamba-2 and Transformer architecture
• 128K context length
• Generates a reasoning trace before producing its final response, unifying reasoning and non-reasoning tasks
• A compact 12B-parameter design that matches or beats leading open models in accuracy while delivering up to 6x higher inference throughput
• Built for enterprise applications that need both accuracy and efficiency

GLM-OCR (Z.ai)

• A 0.9B-parameter multimodal model
• Handles a wide range of inputs: scanned PDFs, handwritten notes, academic papers with equations, multi-column tables, code documents, and multilingual text
• Outputs can be reconstructed as Markdown, JSON, or LaTeX
• Designed for low latency, making it suitable for real-time services and edge devices
• Well suited for large-scale document processing and invoice-extraction workflows

You can deploy any of these models by browsing the model catalog in the SageMaker console, or programmatically via the SageMaker Python SDK.

Wrap-up

• Z.ai's GLM-5.2 FP8 and GLM-OCR, along with NVIDIA's Nemotron-Nano-12B-v2, are now available on Amazon SageMaker JumpStart
• GLM-5.2 FP8 offers a first-ever genuinely usable 1M-token context window for long-horizon agentic software development
• Nemotron-Nano-12B-v2 combines Mamba-2 and Transformer layers, delivering up to 6x higher inference throughput at just 12B parameters
• GLM-OCR is a 0.9B-parameter model that can read everything from scanned documents to handwritten notes and equation-heavy papers

Whether you're building a long-horizon coding agent, need a cost-efficient high-throughput reasoning model, or want to digitize a mountain of documents, this update gives you a new option to try for each use case!