AWS has made Muse Spark Open available through Amazon SageMaker JumpStart
The Muse Spark Open model is available through Amazon SageMaker JumpStart in BF16 and NVFP4 variants. It offers a context window of 1 million tokens and deployment in your own AWS account; billing continues for the entire time it is running.
AWS has made the Muse Spark Open model from Meta available through Amazon SageMaker JumpStart. Users can deploy the BF16 and NVFP4 variants in their own AWS account through Amazon SageMaker Studio or the SageMaker Python SDK. Meta describes the model as optimized for coding and agentic tasks, such as debugging and changes across files.
According to AWS, the model has a mixture-of-experts (MoE) architecture, with 816 billion parameters in total and 42 billion active parameters per token. The context window holds 1 million tokens. The model accepts text, images, documents, audio and video with audio, and returns text. The weights are subject to the Muse Model License, including the Muse Spark Usage Policy. Tools, a browser and an environment for controlling a computer must be provided by the application or agent framework.
The NVFP4 variant runs on an ml.p5e.48xlarge instance, while the BF16 variant runs on ml.p6-b300.48xlarge. According to AWS, the NVFP4 weights require approximately a quarter of the memory required by BF16; even when only some parameters are active, GPU memory must hold all the weights. The deployment supports configuring network isolation, encryption and a pinned model version. The deployed model is billed for the entire time it is running, even without requests.
Why it matters
Developers can try the model on large codebases in their own AWS account. Memory requirements and ongoing billing are key considerations when deciding whether to deploy it: with occasional use, costs accrue even while idle. The availability of the model also allows companies to choose their own network and security settings.
Two audiences, two different impacts
What this means
For individuals
Developers can test how the model handles debugging or changes across files in their project; the focus on these tasks is a claim made by Meta.
For a business
When planning operations, the company must account for a dedicated instance with multiple GPUs and charges even while idle. The choice of variant also determines the required instance type.
DevelopmentCheck the original
Event sources
only one source so far · 1 publisher, 0 independent. We count feeds from the same owner only once.