Run Meta's coding model inside your own AWS account
In brief
According to AWS, Meta has made Muse Spark Open, an open-weight model aimed at software work, available in Amazon SageMaker JumpStart. Meta presents it as a reasoning model with 816 billion parameters in total, 42 billion active per token, and a one-million-token context window. AWS says you can deploy it as an endpoint in your own account, choosing a BF16 version or a more compact NVFP4 version.
Source : AWS Machine Learning Blog - « Meta’s Muse Spark Open now available in Amazon SageMaker JumpStart »
What it changes for an SME
For an SME with developers already on AWS, this offers a coding assistant that, according to AWS, handles your source code and prompts on an endpoint you control. That can lower the risk of sending private repositories to a public service. Cost and operations remain heavy: AWS says the NVFP4 variant runs on ml.p5e.48xlarge and the BF16 variant on ml.p6-b300.48xlarge. Review Meta's Muse Model License and usage policy before any deployment.
What you can do
- Check whether your AWS account has SageMaker access and quota for the GPU instances AWS lists for this model.
- Read the Muse Model License and the Muse Spark Usage Policy before deploying anything.
- If you go ahead, try the NVFP4 variant first, as AWS suggests for cost, and test it on one real internal coding task.
Let's talk about the time you could save.
In 20 minutes, we find where AI could save you time or money. Free, no commitment.