AWS adds serverless customization for two Nemotron 3 tiers

AWS says SageMaker AI's serverless model customization now supports Nemotron 3 Nano and Nemotron 3 Super. Its announcement identifies Nano as 30B total parameters with 3B active, and Super as 120B total parameters with 12B active.

AWS says users can apply supervised fine-tuning, reinforcement learning with verifiable rewards, or reinforcement learning with AI feedback without provisioning or managing infrastructure. It says customization jobs can be started in SageMaker Studio or through the SageMaker Python SDK.

The cited AWS material names Nano and Super, not Ultra, and describes a customization path rather than a change to NVIDIA's underlying weights. For teams already using these models, the update provides an AWS-hosted fine-tuning option, subject to the required AWS permissions, Studio access and prepared training data.