AI CRITIQUE

Bir kahverengi ve bir siyah el, aralarında ahşap bir blok bulunan boş ekranlı bir bilgisayar monitörüne doğru uzanıyor.

Meta's 816-billion-parameter model runs on Mac Studio

Read the article English (automatic translation)

Summary

Meta has released Muse Spark Open, a mixture-of-experts model with 816 billion parameters. With 42 billion active parameters per token, the model runs on two Mac Studio M5 Ultra units via Apple's MLX framework. This development earned rare praise from Apple CEO John Ternus. This collaboration between long-time rivals is positioned less as a product launch and more as a geopolitical signal in the AI race.

Why it matters

This development demonstrates that frontier-scale inference can be performed on consumer-grade desktop hardware without cloud dependency. This strengthens the case for local AI against subscription lock-in. For Apple, it allows the company to value its M5 Ultra and MLX stack as a reliable AI platform without training its own frontier-scale model. This open thaw between Meta and Apple, rivals since the 2021 iOS privacy dispute, indicates that both companies feel intense pressure to avoid being left out of the AI narrative.

Highlights

The mixture-of-experts design (816B total / 42B active), 1 million token context, and local multimodality make the model architecturally ambitious for local deployment.
Apple's own benchmarks: 36.7 tok/s and 4.8s TTFT at 8K context on four M5 Ultras; the Hugging Face card shows a single M3 Ultra 512 GB can run it at approximately 18 tok/s with 2-bit quantization.
Meta has not published any benchmark scores. While describing the model as 'the most capable open-weight model in the US to date,' it referenced an evaluation report that does not yet exist.

Critical take

A model with no published benchmark results, a 'limited commercial' license with retained ownership, and a mandatory hardware configuration (512 GB M5 Ultra) that was not even for sale at the time of launch is a proof of concept dressed up as a product launch. The author's observation that Meta experienced a GitHub repository violation while emphasizing transparency in AI serves as a sharp reminder that the political gesture functions more than the engineering does.

Bu kritik, yerel yapay zeka modeli Qwen3.8-27B tarafından yazılmıştır. Haberi ben seçtim; kritik metnini yayımlanmadan önce tek tek denetlemiyorum.This critique was written by the local AI model Qwen3.8-27B. I chose the story; I do not review each critique before it is published.

Back to the issue