AWS expands SageMaker JumpStart with GLM-5.2, Nemotron-Nano-12B, and gemma-4-12B

AWS expanded its SageMaker JumpStart foundation-model catalog with a broad batch of new open and third-party models, giving AWS customers turnkey deployment access to a wide range of capabilities. The additions include Z.ai's GLM-5.2 FP8 (long-horizon agentic engineering), NVIDIA's Nemotron-Nano-12B-v2 (efficient hybrid reasoning), and GLM-OCR (document understanding), alongside Black Forest Labs' FLUX.2-small-decoder for efficient image generation and Google's gemma-4-12B-it for unified multimodal understanding.

A further tranche added Redis's langcache-embed-v3-small for semantic caching, JetBrains' Mellum2-12B-A2.5B-Thinking for code-focused reasoning, and LightOn's LightOnOCR-2-1B for end-to-end document OCR. The breadth spans agentic engineering, reasoning, OCR, embeddings, and image generation — a deliberate strategy to make SageMaker the neutral one-stop catalog for whatever model a team needs.

By hosting rivals' and partners' models (Google's Gemma, NVIDIA's Nemotron, Chinese GLM) on managed infrastructure, AWS captures the deployment and compute revenue regardless of which model wins on quality. It complements the same-week AgentCore and Bedrock Web Search launches as part of AWS's push to own the enterprise AI control plane.

Skeptics note that catalog breadth doesn't equal differentiation — the models themselves come from others — and the real value is in the managed deployment tooling. What to watch: whether enterprises consolidate on JumpStart versus Bedrock for model access, and how quickly newly-released open models land in the catalog relative to competitors like Azure's model gateway.