Meta’s Muse Spark Open now available in Amazon SageMaker JumpStart

Meta’s Muse Spark Open, a coding‑optimized open‑weight model with 816 billion parameters, has been added to Amazon SageMaker JumpStart. Trained on real developer workflows such as multi‑file refactors and long debugging sessions, the model supports a 1 million‑token context window, enabling it to ingest entire projects, including dependencies and legacy code, without splitting tasks. Two variants are available: a BF16 version (Muse‑Spark‑Open‑816B‑A42B) and an NVFP4 quantized version (Muse‑Spark‑Open‑816B‑A42B‑NVFP4). The NVFP4 variant requires roughly a quarter of the memory of the BF16 weights and runs on ml.p5e.48xlarge, while the BF16 variant runs on ml.p6-b300.48xlarge. Users can deploy the model to a SageMaker AI endpoint in their own AWS account using SageMaker Studio or the SageMaker Python SDK; the deployment process takes 15–20 minutes. The endpoint accepts OpenAI‑compatible chat payloads, returns chat completions with separate reasoning and answer fields, and supports streaming. Deployment options include attaching the endpoint to a VPC, encrypting storage volumes with AWS KMS, and enabling network isolation. Users can delete the endpoint to stop charges. The model is licensed under the Muse Model License, and the Muse Spark Usage Policy applies. Meta describes the model as agentic, capable of planning, acting, observing, and correcting over long horizons, and it can browse the web, coordinate sub‑agents, and operate across desktop, mobile, and browser environments. The model also reasons across images, documents, audio, and video. The article advises benchmarking both variants on user workloads before choosing a production instance.