Back to timeline

Fri, October 1602:15Model/APIModel releasesOpen sourceInfra & costModel releases guide

Meta and Apple partner to make Muse Spark Open run better on Mac

Decision Brief

What changedMeta released the Muse Spark Open open-weights model, optimized for the MLX framework through a partnership with Apple, and it can run locally on two M5 Ultra Mac Studios.
Why it mattersAn 816-billion-parameter, natively multimodal open-weights model running on two Mac Studios means teams wanting local large-model deployment can handle long-horizon inference on desktop hardware.
Who should careTeams building on model APIs
Affected stackNo specific stack identified
Source confidenceMedium · Reliable media or first-hand reporting

Meta today released Muse Spark Open, calling it "the strongest US open-weights model to date." It has 816 billion total parameters, is natively multimodal, and excels at long-horizon tasks, complex software engineering, and multi-step reasoning. Optimized through a partnership with Apple, it can run locally on two Apple M5 Ultra Mac Studios using the latest MLX framework, which Meta calls a first for a model of its class. Meta CEO Mark Zuckerberg and new Apple CEO John Ternus exchanged greetings on X, and Meta Chief AI Officer Alexandr Wang told Axios the two sides have collaborated for months. Apple also published a technical article covering the MLX framework and three open models including Muse Spark Open. For developers and teams deploying local models on Apple Silicon, this means large open-weights models that previously needed server clusters can now be inferred on the desktop via two Mac Studios and MLX, a substantial change in hardware cost and deployment. Teams doing on-device inference with MLX should watch for new framework versions and how to obtain the model weights.

Summary basis: full article readCompiled from the source scope noted above; the original remains authoritative.

Sources

  • The Verge:AI

    Consumer AI products, platform shifts, and policy/culture impact.

  • The Verge:AI

Related intel

留言

登入后即可留言,和其他 builder 交换实测心得。

还没有留言,抢头香。