Muse-Glimmer-30B and Qwen 3.8-27B models now available on Amazon SageMaker JumpStart

Meta’s Muse-Glimmer-30B, a 30-billion-parameter model from the Meta Superintelligence Lab, is designed for autonomous agentic tasks including multi-step reasoning, tool use, and failure recovery. The model integrates a 1.8-billion-parameter ViT-G/14 perception encoder, supports interleaved text and image inputs, and operates with a 131,000+ token context window. Released under the Apache 2.0 license, it can run entirely offline, making it suitable for always-on enterprise agents that require local deployment and resilience to connectivity issues.

Alibaba’s Qwen 3.8-27B is a 27-billion-parameter dense vision-language model optimized for coding, multi-step agentic tasks, and multimodal understanding across text, images, and video. It features a 262,000-token context window, extendable to approximately 1 million tokens via YaRN scaling, and offers adjustable reasoning effort levels. The model achieves a score of 61.7 on SWE-bench Pro and requires around 17 gigabytes when quantized, enabling reliable execution of complex, multi-step workflows in production environments.

Both models are now accessible through Amazon SageMaker JumpStart, allowing customers to deploy them with minimal configuration. Users can access the models via the SageMaker JumpStart model catalog in the SageMaker console or through the SageMaker Python SDK, streamlining integration into existing AWS workflows without requiring extensive setup or infrastructure adjustments.

The addition of Muse-Glimmer-30B and Qwen 3.8-27B expands the range of foundation models available on SageMaker JumpStart, supporting use cases in enterprise automation, multimodal processing, and agentic AI systems. AWS customers can evaluate and deploy these models to address specific AI requirements, leveraging SageMaker’s managed infrastructure for scalable and high-performance inference.