Meta and Apple enable new open-weight Muse model to run on Apple silicon
Meta unveiled Muse Spark Open, an open-weight AI model it says matches top foreign rivals and can run locally on Apple M5 Ultra‑equipped Macs under a new MLX workflow.

Meta unveiled Muse Spark Open, an open-weight AI model it says matches top foreign rivals and can run locally on Apple M5 Ultra‑equipped Macs under a new MLX workflow.

President Trump said the FBI is investigating an alleged relationship between Nvidia chief Jensen Huang and a Chinese national accused of spying. Trump said he spoke with Huang, who denied knowing the person was a spy.
4 sources
Nvidia's chief executive abruptly left his post after reports that one of his extramarital relationships involved a woman suspected of being a Chinese intelligence operative; he has been summoned to testify before Congress, according to one outlet.
4 sources
Published 19 Oct 2026
Meta on Thursday introduced Muse Spark Open, a new open‑weight artificial intelligence model the company says competes with leading offerings from Chinese developers and is engineered to run locally on Apple silicon.
The company describes the model as capable of advanced multi‑step reasoning and handling complex software engineering tasks, according to Meta’s business statement. Alexandr Wang, Meta’s head of AI, posted about the launch on X, calling the release a milestone for US open‑weight models.
Meta and Apple worked together to optimise the model for Apple's latest chips, with several outlets reporting the model is tuned for the M5 Ultra. TechCrunch and Punch said the model can run on Apple machines locally; The Verge added that Muse Spark Open is the first model in its category that Meta says can operate on a desktop setup using two M5 Ultra Mac Studio units together with the company’s newest MLX framework.
Meta framed Muse Spark Open as an attempt to narrow gaps with competitors abroad. In its announcement, the firm highlighted the model’s strengths in multi‑step reasoning and software engineering tasks — capabilities it positioned as part of its push to offer high‑performance open‑weight options from US developers.
Open‑weight models are distributed in ways that let third parties inspect or run the base model weights, which can speed research and product development compared with closed commercial models. Meta’s emphasis on making the model runnable on personal desktops aims to expand who can experiment with and deploy advanced AI without exclusive access to vast cloud resources.
Optimising for Apple’s M5 Ultra chips and providing a workflow that can run on local Mac hardware could lower the hardware barrier for organisations and individual developers that use Apple machines. The Verge’s description of a two‑Mac Studio setup using the MLX framework suggests Meta expects some deployments to rely on substantial local compute, even if that makes desktop‑based experimentation feasible.
Outlets reporting the launch mostly repeat Meta’s claims about the model’s capabilities and its relative standing against competitors; independent benchmarks or third‑party evaluations have not been cited by the reports. Details such as licensing, exact performance comparisons, and wider hardware requirements beyond the Apple M5 Ultra configuration were not provided in the initial coverage.
Meta’s move follows industry trends toward more portable, locally runnable models and closer hardware partnerships. By promoting an open‑weight model that runs on popular consumer silicon, Meta and Apple are signalling a strategy that combines accessible model distribution with vendor‑specific optimisation.
This story was written from reports by these outlets. Read the originals:
Nvidia says its board is investigating whether CEO Jensen Huang was the target of a foreign intelligence effort after he acknowledged an affair with a Chinese businesswoman whom U.S. reports have linked to Beijing.
9 sources
Nvidia said co‑founder Jensen Huang has resigned as CEO and board member effective immediately following reports he had multiple extramarital relationships, including one linked by reports to a suspected Chinese agent.
5 sources