Meta releases Muse Spark Open, an 816B-parameter model
The open-weight release of Meta's Muse Spark model ships with 42B active parameters, a 1M-token context window and a free licence for most commercial use.
Meta released open weights for Muse Spark Open on 15 October, calling it the most capable open-weight model from a US lab to date. It is natively multimodal, with 816 billion total parameters and 42 billion active per forward pass, and a 1M-token context window.
Meta built the model for coding and agentic work: long-horizon tasks, complex software engineering and multi-step reasoning, which it demonstrates through iOS development and video-editing examples. Weights are on Hugging Face, Amazon SageMaker AI, Google Cloud Model Garden and Microsoft Foundry Models, and the model runs on SGLang, vLLM, TensorRT-LLM, Transformers, Apple MLX, llama.cpp and Unsloth. Meta also published custom Helion GPU kernels for SGLang that it says give a 1.03 to 3.95 times kernel-level speedup on Nvidia and AMD GPUs.
Why it matters: this is a frontier-scale model you can download and run yourself rather than only call through an API. Meta licenses it under its Muse Model License with, in Meta’s words, “limited commercial terms” that let most developers use it for free, so check the licence before building a product on top of it.
The caveat: running the full model yourself takes serious hardware. Meta’s own example setup is two Mac Studios, each with 512GB of unified memory.