Apple and Meta bury the hatchet so Muse Spark Open can run on your desk

What's the deal? Meta has released Muse Spark Open, which it calls the most capable US open-weight model to date, and partnered with longtime rival Apple to make it the first model in its class engineered to run locally on consumer hardware — two to four Apple M5 Ultra Mac Studios running Apple's MLX framework, with roughly 1TB of combined unified memory.

The specs: Muse Spark Open is a natively multimodal mixture-of-experts model with 816 billion total parameters, 42 billion active, and a 1-million-token context window, built for coding and long-horizon agentic work. It is distinct from Muse Wonder, Meta's proprietary frontier model released earlier this week.

Why Mac? Apple's unified memory architecture lets each M5 Ultra's GPU address up to 512GB, so two to four Mac Studios can hold the entire model between them. Meta says the linked machines "behave like one supercomputer with a terabyte of unified memory."

Why now? The release ends years of public friction between Apple and Meta over privacy and app-store policy. Axios quoted Scale AI's Alexandr Wang saying he was "excited for mommy and daddy to be playing nice again," while Apple's John Ternus and Meta's Mark Zuckerberg traded congratulatory posts on X. The tie-up gives Meta a flagship consumer showcase for its open models and gives Apple a marquee AI use case for its newest silicon.

The fine print: Weights are live today on Hugging Face under the Muse Model License, which carries limited commercial terms. The model is also available through API platforms including BaseTen, Fireworks AI, Groq and Together AI.

The signal: A rare alliance between two Silicon Valley rivals pushes a frontier-scale open-weight model out of the data centre and onto the desktop — a preview of locally-run AI that doesn't depend on the cloud.

Read more: Meta AI Research · Axios · CNBC

Image credit: Yasu, CC BY-SA 3.0, via Wikimedia Commons