Meta releases Muse Spark Open weights: 816B MoE, 1M context, runs on two Mac Studios
Meta open-sourced the weights of Muse Spark Open, a multimodal coding and agent model it calls the most capable US open-weight model to date.
Muse Spark Open is a natively multimodal model with 816B total parameters (42B active) and a 1M-token context window, aimed at coding, long-horizon agent tasks and multi-step reasoning.
- Where to get it: weights on Hugging Face, Amazon SageMaker AI, Google Cloud Model Garden and Microsoft Foundry. Hosted APIs at Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter and Together AI.
- Runs on: SGLang, vLLM, TensorRT-LLM, Transformers, MLX, llama.cpp and Unsloth.
- Local: Meta says it’s the first model in its class to run on a desktop, using two Apple M5 Ultra Mac Studios with MLX.
The “most capable” claim is Meta’s own, so wait for independent benchmarks.
Why it matters
It's a frontier-class open model you can self-host or fine-tune, and Meta says it can run locally on two Apple M5 Ultra Mac Studios. It's also on hosted APIs like Groq, Together, Fireworks and OpenRouter on day one.