Apple and Meta team up to run Muse Spark Open on Macs
Apple and Meta collaborated to help developers run Meta’s Muse Spark Open locally on high-end Mac computers. Meta describes the model as its most capable U.S. open-weight model to date.

Key takeaways · 4
- 01
The collaboration targets local use of Muse Spark Open on high-end Macs.
- 02
Meta says the model has a 1-million-token context window and takes text, image, audio and video input, with text output.
- 03
Meta says two M5 Ultra Mac Studios and MLX can run the model on a desktop.
- 04
Local use may add security and save money compared with cloud use, according to the companies.
What the companies are enabling
Apple and Meta collaborated to help developers run Muse Spark Open locally on high-end Mac computers.[1] The model is an open-weight release: developers can download its parameters and run it on different hardware.[1] Meta CEO Mark Zuckerberg said the weights were releasing that day and that the model ran well on new Mac Studios after work with Apple and John Ternus.[1] Ternus said he was amazed that a model of this caliber and scale could run efficiently from a desk.[1]
Model capabilities and workload
Meta says Muse Spark Open has 816 billion total parameters, of which 42 billion are active, and a context window of 1 million tokens.[2] It accepts text, image, audio and video input and produces text output.[3] Meta says it is built for coding and agentic work, including long-horizon tasks, complex software engineering and multi-step reasoning.[2] The model routes each token to a small subset of experts, and Meta says per-token compute tracks the 42 billion active parameters.[3]
Hardware and access
Meta says a desktop can run Muse Spark Open using two Apple M5 Ultra Mac Studios and the MLX framework.[2] CNBC reported that customers can pair two to four Mac Studio computers to use the model.[1] The hardware requirement has a substantial memory implication: Meta says serving the model still requires loading all 816 billion weights into GPU memory.[3] CNBC reported that a Mac Studio with the required M5 Ultra chip and 256GB of memory costs at least $10,799, and that Apple planned a more expensive 512GB RAM version later that month.[1]
Local use, tests and limitations
The companies said running the model locally rather than in the cloud adds security and can save users money.[1] Meta says its on-device tests included using the model to ship iOS changes to the Ice Cubes app and to review footage and crop it vertically in DaVinci Resolve.[2] Developers can also access the model through listed API platforms, while its weights are available through several cloud and model-hosting services.[2] Meta released the weights under its Muse Model License, which it says allows many developers to use the model for free under limited commercial terms.[2] Axios noted that open-weight models raise concerns about bad actors removing safeguards for cyberattacks or biological threats.[4]
For engineering teams, the announcement offers another route to experiment with a large model on local hardware, but the total-weight memory requirement and Mac cost are material planning constraints. Teams considering use should weigh the local-workflow benefits against hardware needs, license terms and the risks associated with open weights.
Why it matters
Test yourself on this story — 1 question.
Create a free account to take the quiz, earn XP, and get a daily session built for your industry.
Take the quizHow this developed
15 October 2026
Apple and Meta team up to run Muse Spark Open on Macs