Meta launches an open-weight AI model that runs on Macs
Meta launched Muse Spark Open, an 816-billion-parameter open-weight model with a 1 million token context window that runs locally on Apple silicon Macs.
By Dillip Chowdary • Oct 15, 2026 • Source: TechCrunch
At a glance
- Model
- Muse Spark Open
- Parameters
- 816 billion
- Context window
- 1 million tokens
- Local RAM
- at least 1 terabyte (per Engadget)
- Best hardware
- Mac Studio with M5 Ultra chip
- Availability
- now, via several API platforms
In this post
Meta has launched Muse Spark Open, an open-weight model it calls the "most capable" in the United States, designed to run on Macs. Meta and Apple teamed up so the newest model runs locally on Apple silicon, according to TechCrunch's report.
It is natively multimodal. On Thursday Meta released evaluation results that claim superior performance to other popular open-weight models.
This piece covers the model, the Mac a local run expects, and the desktop jobs Meta has already run. It is for engineers on an AI coding terminal, teams running agentic workflows, and editors in DaVinci Resolve, plus developers who will call an API when a computer cannot hold the weights. Local use requires at least 1 terabyte of RAM.
01The announcement
Meta and Apple teamed up on the launch, and the local version is built for Apple silicon Macs. Muse Spark Open is natively multimodal. Meta is making the model available now through API platforms and on Macs that meet the memory requirement for a local run.
Meta and Apple teamed up on the launch, and the local version is built for Apple silicon Macs. Muse Spark Open is natively multimodal. Meta is making the model available now through API platforms and on Macs that meet the memory requirement for a local run.
The published scale is 816 billion parameters and a 1 million token context window. Meta released evaluation results on Thursday and said the model offers superior performance to other popular open-weight models.
Those Thursday results sit beside the parameter count and the context length as the backing for that claim. Meta built Muse Spark Open for complex coding tasks and agentic labor, and designed it to run locally on Apple's computers when the hardware can carry a model of this size.
02What actually changed with Meta an open-weight AI model that runs
Meta paired Muse Spark Open with Muse Code, its AI coding terminal, and deployed the combination across various agentic workflows. The coding example was a Mastodon client.
Meta said that work included "bug fixes, new features, [and] full build." The phrase covers fixes, new features, and a completed build, produced through the coding terminal. In that launch, Muse Spark Open sits in Muse Code and has already been run on that Mastodon client for bug fixes, new features, and a full build.
Meta used Muse Spark Open to run DaVinci Resolve, the high-end video editing software, and said the model reviewed footage and cropped it. Review and crop are the tasks Meta reported.
The model is built to run locally on Apple's computers and runs best on the Mac Studio equipped with the M5 Ultra chip. Meta says open-weight models can be significantly more cost-efficient than closed and proprietary ones, and can offer transparency and flexibility that closed source models do not allow.
Advertisement
Tech Pulse Daily
Get tomorrow's pulse first
Join engineers who read Tech Pulse before stand-up. Free, weekday mornings.
03Who should care
Running Muse Spark Open locally takes at least 1 terabyte of RAM, per Engadget, and still takes a lot of power. The best fit is the Mac Studio with the M5 Ultra chip, so owners of that Apple silicon studio can run the model on the machine.
Coding teams are the other audience. Meta put the model in Muse Code, and the Mastodon client work (bug fixes, new features, and a full build) is the job the company says it already ran.
Developers under the 1 terabyte RAM line have the API route: Muse Spark Open is available now through a number of different API platforms. Meta says open-weight models can be significantly more cost-efficient and can offer transparency and flexibility that closed and proprietary models do not allow. It calls Muse Spark Open the "most capable" open-weight model in the United States.
Developers under the 1 terabyte RAM line have the API route: Muse Spark Open is available now through a number of different API platforms. Meta says open-weight models can be significantly more cost-efficient and can offer transparency and flexibility that closed and proprietary models do not allow. It calls Muse Spark Open the "most capable" open-weight model in the United States.
04How to try it
On a Mac, a local run needs at least 1 terabyte of RAM, per Engadget, and the model runs best on the Mac Studio equipped with the M5 Ultra chip. Apple silicon is the hardware in the Meta and Apple pairing. A local run still needs a lot of power.
On a Mac, a local run needs at least 1 terabyte of RAM, per Engadget, and the model runs best on the Mac Studio equipped with the M5 Ultra chip. Apple silicon is the hardware in the Meta and Apple pairing. A local run still needs a lot of power.
Muse Code is Meta's AI coding terminal. Meta paired it with Muse Spark Open for agentic workflows.
Video follows the DaVinci Resolve deployment, where the model reviewed and cropped footage. A local trial still depends on memory. Engadget puts the minimum RAM for a computer run at 1 terabyte.
Video follows the DaVinci Resolve deployment, where the model reviewed and cropped footage. A local trial still depends on memory. Engadget puts the minimum RAM for a computer run at 1 terabyte.
05What to watch next
The specs with that claim are 816 billion parameters and a 1 million token context window. Meta calls the model the "most capable" open-weight model in the United States and says it is natively multimodal, built for complex coding tasks and agentic labor. Those Thursday results, the parameter count, the context length, and that ranking are the performance details shipped with the launch.
The specs with that claim are 816 billion parameters and a 1 million token context window. Meta calls the model the "most capable" open-weight model in the United States and says it is natively multimodal, built for complex coding tasks and agentic labor. Those Thursday results, the parameter count, the context length, and that ranking are the performance details shipped with the launch.
A local run needs at least 1 terabyte of RAM, per Engadget, and runs best on a Mac Studio with the M5 Ultra chip on Apple silicon. Computers under that RAM figure use the API path, where Muse Spark Open is available now on a number of different platforms.
With Muse Code, a Mastodon client got "bug fixes, new features, [and] full build." In DaVinci Resolve the model reviewed and cropped footage. Those runs sit next to Thursday's performance claim.
Frequently asked questions
What is Muse Spark Open?
Muse Spark Open is the open-weight model Meta launched and calls the "most capable" in the United States. It has 816 billion parameters, a 1 million token context window, and it is natively multimodal. Meta and Apple teamed up so it runs locally on Apple silicon.
How much RAM does Muse Spark Open need on a Mac?
A local run needs at least 1 terabyte of RAM, per Engadget, and still needs a lot of power. The model runs best on a Mac Studio equipped with the M5 Ultra chip.
Where is Muse Spark Open available?
Meta is making Muse Spark Open available now through a number of different API platforms. Local use is for Apple silicon Macs that meet the 1 terabyte memory requirement.
What is Muse Code?
Muse Code is Meta's AI coding terminal. Meta paired Muse Spark Open with Muse Code for agentic workflows, including a Mastodon client whose work Meta described as "bug fixes, new features, [and] full build."
What did Muse Spark Open do in DaVinci Resolve?
Meta deployed the model on DaVinci Resolve, the high-end video editing software, and said it reviewed footage and cropped it.
Sources
Author
Dillip Chowdary
Writes Tech Bytes coverage of AI, engineering, and the tools that actually ship. Editor of Tech Pulse Daily.
Related on Tech Bytes
Advertisement