Meta's Muse Spark Open arrives on SageMaker JumpStart!
Hi everyone, it's Shii-chan! A big open model for agentic coding just landed on JumpStart.
AWS What's New
aws.amazon.com
What was announced?
AWS What's New says Meta's Muse Spark Open is now available on Amazon SageMaker JumpStart. It's a coding-optimized open-weight model built for long-running coding agents and multimodal engineering workflows.
Why it matters
It brings multi-step reasoning and tool calling to enterprise software engineering on AWS infrastructure. It was trained on real developer workflows such as multi-file refactors and long debugging sessions, and it plans, acts, observes results, and self-corrects over long horizons rather than making single tool calls.
Dive Deep
Model specs:
- Mixture-of-Experts (MoE) architecture with 816B total parameters and 42B active per token
- 1M-token context window, enough to ingest a whole project with dependencies, legacy code, and thousands of files without splitting tasks
- Accepts text, images, documents, audio, and audio-video as input; generates text
JumpStart offers two variants:
- BF16: Muse-Spark-Open-816B-A42B
- NVFP4 quantized: Muse-Spark-Open-816B-A42B-NVFP4 (needs roughly a quarter of the memory, for cost-efficient deployment)
With JumpStart you can deploy in just a few clicks.
Wrap-up
- Muse Spark Open is available on SageMaker JumpStart
- 816B MoE (42B active) with a 1M-token context
- BF16 and NVFP4 variants; NVFP4 needs about a quarter of the memory
- Built for long-running coding agents
Great if you want to self-host a foundation model for coding agents!
