Meta Releases Muse Spark Open, an 816B Open-Weight Model Built to Run on Two Mac Studios

The multimodal Mixture-of-Experts model has 42 billion active parameters and a 1 million-token context window. Weights are live on Hugging Face, and Meta says it is the most capable U.S. open-weight model to date.

MENLO PARK, Calif. - Meta on Thursday released the weights for Muse Spark Open, a natively multimodal model it calls the most capable U.S. open-weight system to date. The model has 816 billion total parameters with 42 billion active, a 1 million-token context window, and is aimed at coding and agentic work such as long-horizon tasks, complex software engineering and multi-step reasoning.

“Today we’re releasing Muse Spark Open, the most capable US open-weight model to date,” Meta’s @AIatMeta account wrote on X. “Built for coding and agentic work, it excels at long-horizon tasks, complex software engineering, and multi-step reasoning.”

Screenshot: @AIatMeta on X, Oct. 15, 2026

The desktop pitch is the part that broke through. Meta says Muse Spark Open is the first model in its class to run on two Apple M5 Ultra Mac Studios using the latest MLX framework. Each Studio can hold up to 512 GB of unified memory for the M5 Ultra’s integrated GPU, so a pair behaves like a machine with a terabyte of memory and 1.2 TB/s of bandwidth. Mark Zuckerberg posted that the model “runs great on new Mac Studios thanks to work with @johnternus and Apple,” and Apple hardware chief John Ternus replied that it was “amazing to run a model of Muse Spark Open’s caliber and scale this efficiently, right from your desk.”

Mac Studio on a desk, from Apple’s M5 Max and M5 Ultra announcement. Photo: Apple

Screenshot image from Mark Zuckerberg’s post showing Mac Studios. Image: @finkd on X

Alexandr Wang, who leads Meta Superintelligence Labs, framed the release the same way. “1/ muse spark open is the most capable US open-weight model to date,” he wrote. “816B total params, 42B active. 1M-token context. natively multimodal. built for coding and agentic work.” He later posted a short film under the line “why 2026 won’t be like 1984,” nodding at Apple’s famous desk-bound computing pitch.

Screenshot: @alexandr_wang on X, Oct. 15, 2026

Meta says it paired the model with Muse Code, its coding terminal, and ran agentic workflows entirely on-device. One demo shipped real iOS changes to Ice Cubes, an open-source Mastodon client, covering bug fixes, new features and a full build. Another drove DaVinci Resolve end to end, with the model reviewing footage frame by frame to find strong moments and crop them vertically.

Muse Spark Open artwork shared with Wang’s thread. Image: @alexandr_wang / Meta

Mac Studio running creative apps including DaVinci Resolve. Photo: Apple

Weights are available now on Hugging Face, Amazon SageMaker AI, Google Cloud Model Garden and Microsoft Foundry Models. Hosted inference is live through Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter and Together AI. Meta says the open-weight build ships with custom Helion GPU kernels in SGLang that deliver 1.03x to 3.95x kernel-level speedups on NVIDIA and AMD GPUs, and a 3 percent end-to-end speedup on AMD’s MI350. Developers can further tune the model with PyTorch’s TorchTitan. The company is releasing the weights under the Muse Model License with limited commercial terms that allow many developers to use the model for free, and it says Muse Spark Open received additional safety post-training and met the thresholds in Meta’s Superintelligence Scaling Framework for release.

Screenshot: @finkd on X, Oct. 15, 2026

Video: Alexandr Wang’s Muse Spark Open desk film, Oct. 15, 2026

Sources: Meta AI Research, “Introducing Muse Spark Open, The Most Capable US Open-Weight Model to Date,” Oct. 15, 2026, https://research.meta.ai/blog/muse-spark-open (816B/42B active, 1M context, multimodal, Mac Studio MLX, Ice Cubes and DaVinci Resolve demos, API partners, weight hosts, Helion kernels, Muse Model License, safety framework); @AIatMeta on X, Oct. 15, 2026, https://x.com/AIatMeta/status/2110779754274292212; @alexandr_wang on X, Oct. 15, 2026, https://x.com/alexandr_wang/status/2110778736434184611 and https://x.com/alexandr_wang/status/2110780784051474666; @finkd on X, Oct. 15, 2026, https://x.com/finkd/status/2110778025629347999; @johnternus on X, Oct. 15, 2026, https://x.com/johnternus/status/2110789142326165518; Hugging Face, meta-models/muse-spark-open collection; TechCrunch, Lucas Ropek, Oct. 15, 2026 (Mac Studio memory context).

Jordan Hale, AI. Jordan Hale is a senior correspondent covering artificial intelligence, chips, and lab strategy. Previously covered enterprise tech for national outlets; based in New York.