Amazon SageMaker AI now supports serverless model customization for Muse Glimmer, Qwen 3.8 and Ministral 3 models

Amazon SageMaker AI now allows serverless customization of Muse Glimmer, Qwen3.8 27B, and Ministral 3 models, expanding the range of foundation models available for adaptation. Users can apply supervised fine-tuning, direct preference optimization, and reinforcement fine-tuning to tailor these models to specific domains and workflows. The update also supports low-rank adaptation and full fine-tuning methods for precise model adjustments. This addition joins existing serverless customization options for models from Nova, Gemma 4, Nemotron 3.5, Qwen, Llama, gpt-oss, and DeepSeek families.

The serverless approach automates infrastructure provisioning and training orchestration, letting users focus on data and evaluation rather than managing clusters. Organizations can use labeled data with supervised fine-tuning to improve accuracy on domain-specific tasks or preference data with direct preference optimization to align outputs with their tone. Reinforcement fine-tuning allows the use of reward signals to enhance performance on new tasks, all while reducing operational overhead.

Serverless model customization is available in US East (N. Virginia), US West (Oregon), Asia Pacific (Tokyo), and Europe (Ireland), providing regional flexibility for deployment. Users can initiate customization jobs through the Models page in Amazon SageMaker Studio, the SageMaker Python SDK, or Agent skills with coding assistants. The process requires no manual infrastructure setup, shifting costs to usage-based billing for training and adaptation tasks.

The update aims to help organizations deploy smaller, cost-efficient models that match the quality of larger frontier models on specific tasks. By leveraging proprietary data, users can reduce latency and expenses while maintaining high performance. Documentation and supported model lists are available to guide users through the customization process and available options.