Meta releases Muse Spark Open, an 816B open-weight model

What happened: Meta has released Muse Spark Open, a natively multimodal open-weight model. Meta AI Research calls it "the most capable US open-weight model to date."

The numbers: It has 816B total parameters, with 42B active in a mixture-of-experts design. The context window is 1M tokens.

The focus: Meta is aiming at coding and agentic work: long-horizon tasks, complex software engineering and multi-step reasoning.

Run it locally: Two Apple M5 Ultra Mac Studios can run it via MLX. Each offers up to 512 GB of unified memory and 1.2 TB/s of bandwidth. Together they act like one machine with about 1 TB of memory.

Where to get it: Weights are on Hugging Face, Amazon SageMaker AI, Google Cloud Model Garden and Microsoft Foundry Models. Hosted APIs come from Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter and Together AI.

The details: Supported engines include SGLang, vLLM, TensorRT-LLM, Transformers, MLX, llama.cpp and Unsloth. Helion kernels in SGLang deliver 1.03–3.95x kernel speedups on Nvidia and AMD GPUs.

The licence: Weights ship under the Muse Model License, which has limited commercial terms. Many developers can use the model free of charge.

The catch: The blog's benchmark section gave no figures in the extracted text. The performance claims await independent testing.

Why it matters: Meta says the model received extra safety post-training and met the release thresholds in its Meta Superintelligence Scaling Framework.

What's next: Meta says the full evaluation report will be "available soon."