Meta and Apple team up to make Muse AI models run better on Macs

The source is short, so I’m drafting from only those names, numbers, and quotes, then checking each section’s length.Meta announced the Muse Spark Open AI model on October 15, 2026, calling it the most capable US open-weight model to date and the first model in its class that can run on a desktop using two Apple M5 Ultra Mac Studios with the latest MLX framework. Meta and Apple, typically rivals, worked together on the launch and are publicly supporting that work. The Verge's report by Jay Peters, time-stamped 6:15 PM UTC, carries Zuckerberg's note that the weights are releasing today and John Ternus's reply about running the model from a desk.

This piece covers Meta's description of Muse Spark Open, the 816 billion parameters and the tasks Meta named, the two-Mac-Studio MLX setup, and the public comments from Mark Zuckerberg, John Ternus, and Alexandr Wang. It is for engineers weighing an open-weight model aimed at complex software engineering and multi-step reasoning, and for readers following a Meta-Apple collaboration on desktop AI.

Meta and Apple team up to make Muse AI models: the announcement

Meta CEO Mark Zuckerberg wrote on X, "Muse Spark Open -- weights releasing today -- runs great on new Mac Studios thanks to work with @johnternus and Apple." John Ternus, the new Apple CEO, replied, "Congrats to the team!" He also wrote, "Amazing to run a model of Muse Spark Open's caliber and scale this efficiently, right from your desk. Mac Studio keeps pushing the boundaries of on-device AI!" The two posts are the CEOs' public support for each company's work on Muse Spark Open.

Meta chief AI officer Alexandr Wang told Axios, "We're excited for mommy and daddy to be playing nice again." Axios noted that Meta and Apple have been collaborating over several months. Jay Peters, the senior reporter covering technology, gaming, and more who wrote the account, joined The Verge in 2019 after nearly two years at Techmeme. The photo, by Amelia Holowaty Krales / The Verge, carries the caption that the new Muse Spark Open AI model can work on M5 Ultra Mac Studios.

What actually changed with Meta and Apple team up to make Muse AI models

Meta described Muse Spark Open as a natively multimodal model with 816 billion total parameters that excels at long-horizon tasks, complex software engineering, and multi-step reasoning. Meta also called it the most capable US open-weight model to date and the first model in its class that can run on a desktop using two Apple M5 Ultra Mac Studios with the latest MLX framework. Zuckerberg said on X that the weights were releasing the day of the announcement, October 15, 2026.

Apple published a new technical post about MLX, an array framework for machine-learning tasks on Apple Silicon, together with Muse Spark Open and two other open models. That post is Apple's write-up of the framework Meta named for the desktop run. The hardware in Meta's claim is two Apple M5 Ultra Mac Studios, and the photo caption states that the new Muse Spark Open AI model can work on M5 Ultra Mac Studios. MLX is the latest framework Meta paired with that pair of machines.

Who should care about Meta and Apple team up to make Muse AI models

Meta's task list names the work this model is pitched for: long-horizon tasks, complex software engineering, and multi-step reasoning, on a natively multimodal model with 816 billion total parameters. Meta's ranking line for that audience is "most capable US open-weight model to date." The open-weight part is tied to Zuckerberg's statement that the weights are releasing the day of the announcement, so engineers doing that complex software and reasoning work have the weight release Mark Zuckerberg placed on October 15, 2026.

The hardware audience is anyone running a desktop of two Apple M5 Ultra Mac Studios with the latest MLX framework, the setup Meta called the first in its class. Ternus aimed his reply at that desk, writing that it is amazing to run a model of Muse Spark Open's caliber and scale this efficiently, right from your desk, and that Mac Studio keeps pushing the boundaries of on-device AI. Readers tracking the rivalry have Wang's Axios comment and the report that the companies collaborated over several months.

How to try Meta and Apple team up to make Muse AI models

Zuckerberg posted the availability line on X: "Muse Spark Open -- weights releasing today -- runs great on new Mac Studios thanks to work with @johnternus and Apple." Today, in that post, is October 15, 2026. Meta's run description says Muse Spark Open is the first model in its class that can run on your desktop using two Apple M5 Ultra Mac Studios with the latest MLX framework. The stated machine count is two M5 Ultra Mac Studios, and the stated software layer is the latest MLX.

Apple's published path alongside those weights is a new technical post about MLX. The post treats MLX as an array framework for machine-learning tasks on Apple Silicon and discusses Muse Spark Open plus two other open models. The Verge account leaves those two models unnamed. What it specifies for a desktop try is the October 15, 2026 weight release, two Apple M5 Ultra Mac Studios, and the latest MLX framework that Meta credited for running Muse Spark Open on that desk.

What to watch after Meta and Apple team up to make Muse AI models

Axios said the companies have been collaborating over several months, and the launch-day record is a pair of posts. Zuckerberg thanked work with @johnternus and Apple and said the weights were releasing that day. Ternus answered "Congrats to the team!" and said it is amazing to run a model of Muse Spark Open's caliber and scale this efficiently, right from your desk. Wang's line to Axios — "We're excited for mommy and daddy to be playing nice again" — is how Meta's chief AI officer described the companies working together.

Apple's MLX technical post covers the array framework, Muse Spark Open, and two other open models that the story does not name. The Meta claims sitting next to the released weights are the 816 billion total parameters, the natively multimodal design, and the stated strength on long-horizon tasks, complex software engineering, and multi-step reasoning. The same claims include "most capable US open-weight model to date" and "the first model in its class" on two Apple M5 Ultra Mac Studios with the latest MLX framework.