August 11, 2026, (Inside AI) — Amazon Web Services has added three new foundation models to its SageMaker JumpStart service, giving enterprise customers one-click access to Z.ai’s GLM-5.2 FP8, NVIDIA’s Nemotron-Nano-12B-v2, and Z.ai’s GLM-OCR. The deployment, announced today, targets specialized use cases in agentic engineering, hybrid reasoning, and document understanding.
The trio represents a strategic expansion of AWS’s managed model catalog, which now includes over 80 pre-trained models. Each new entrant addresses a distinct bottleneck in enterprise AI adoption: GLM-5.2 FP8 tackles long-horizon software development tasks, Nemotron-Nano-12B-v2 focuses on inference efficiency for reasoning workloads, and GLM-OCR delivers compact document parsing for real-time services.
GLM-5.2 FP8 is specifically optimized for agentic engineering workflows. It can manage full-cycle software development, from requirements gathering to deployment, within a single task. For the first time in the GLM series, the model offers a 1M-token context window, enabling it to process project-level engineering context and maintain consistency across long-running operations. This marks a significant leap over its predecessor, GLM-5.1, which struggled with extended task reliability.
NVIDIA’s Nemotron-Nano-12B-v2 brings a hybrid architecture to the table, combining Mamba-2 and Transformer components. With a 128K context length and 12B parameters, it generates reasoning traces before final answers, delivering up to 6x higher inference throughput than comparable open models while maintaining accuracy. This makes it suitable for enterprise applications that demand both speed and precision in unified reasoning and non-reasoning tasks.
Read: OpenAI Privacy-Filter for PII Detection Now on AWS SageMaker JumpStart
GLM-OCR is a 0.9B-parameter multimodal model designed for document understanding. It processes scanned PDFs, handwritten notes, academic papers with formulas, multi-column tables, code documentation, and multilingual text, reconstructing them into clean Markdown, JSON, or LaTeX. Its low latency suits real-time services and edge devices, targeting large-scale document processing and invoice extraction.
“Z.ai’s GLM-5.2 FP8, NVIDIA’s Nemotron-Nano-12B-v2, and Z.ai’s GLM-OCR models are now available on Amazon SageMaker JumpStart, expanding the portfolio of foundation models available to AWS customers,” stated the AWS announcement.
The move intensifies competition in the managed AI model market. AWS, Microsoft Azure, and Google Cloud have all been racing to onboard third-party models. SageMaker JumpStart’s one-click deployment lowers the barrier for enterprises to experiment with specialized models without managing infrastructure. However, some industry observers note that the real test will be how these models perform on domain-specific benchmarks, not just generic leaderboards.
NVIDIA’s Nemotron-Nano-12B-v2, for instance, enters a crowded field of small language models optimized for inference. Its hybrid architecture is reminiscent of recent research into state-space models, but its practical advantage over pure Transformer models in production environments remains to be seen. Meanwhile, GLM-5.2 FP8’s 1M-token window pushes the boundaries of long-context processing, an area where models like Gemini have also made strides.
GLM-OCR’s 0.9B parameter count is notably compact for a multimodal model, potentially enabling on-device deployment. This aligns with a broader industry trend toward smaller, specialized models that can run locally, reducing latency and privacy concerns. AWS’s integration could accelerate adoption in sectors like legal, finance, and healthcare where document processing is critical.
Customers can access the models through the SageMaker JumpStart model catalog in the AWS console or via the SageMaker Python SDK. AWS has not disclosed pricing details, but costs will likely depend on the underlying instance types and inference duration. For more information, see the Amazon SageMaker JumpStart documentation.
Read: OpenAI GPT-5.6 Models on AWS Bedrock Now Support 1M Token Context
If a research paper dropped this week, Mahesh has already read it. He covers AI tools and vibe coding for builders who need to know what is worth their time and what is not.
