Muse Spark Open model now available on Amazon SageMaker JumpStart

Meta’s Muse Spark Open, an open-weight model designed for coding tasks, is now available through Amazon SageMaker JumpStart. This integration allows AWS customers to leverage the model for agentic coding workflows, including multi-step reasoning and tool calling, directly within their AWS infrastructure. The model is optimized for long-running coding agents and multimodal engineering tasks, supporting inputs such as text, images, documents, audio, and audio-video while generating text outputs.

Muse Spark Open is built on a Mixture-of-Experts architecture with 816 billion total parameters and 42 billion active per token, balancing high capacity with efficient compute. Its 1 million-token context window enables it to process entire projects—including dependencies and legacy code—without task segmentation. The model is trained on real developer workflows like multi-file refactors and long debugging sessions, allowing it to plan, act, observe results, and self-correct over extended periods.

Amazon SageMaker JumpStart offers two variants of Muse Spark Open: BF16 (Muse-Spark-Open-816B-A42B) and NVFP4 quantized (Muse-Spark-Open-816B-A42B-NVFP4). The NVFP4 variant reduces memory usage by approximately 75%, making it more cost-efficient for deployment while maintaining performance. Customers can deploy either variant with minimal effort, as SageMaker JumpStart handles the infrastructure setup.

Deployment is streamlined through the SageMaker JumpStart model catalog in the SageMaker console or via the SageMaker Python SDK. Endpoints are created within the customer’s AWS account, ensuring full control over source code and prompts. For additional guidance, AWS provides documentation on deploying and using foundation models in SageMaker JumpStart.