Meta's Muse Spark Open agentic coding model lands on AWS SageMaker JumpStart

Meta's open-weight Muse Spark Open model is now available in Amazon SageMaker JumpStart. The MoE model has 816B total parameters with 42B active per token and a 1M-token context window, and ships in BF16 and NVFP4 quantized variants.

• MoE architecture: 816B total parameters, 42B active per token
• 1M-token context window handles entire projects and legacy code
• Two variants: BF16 and NVFP4 quantized with ~4x less memory
• Deploys in the customer's own AWS account via SageMaker JumpStart