Meta Releases Muse Spark Open, an 800B-Parameter AI Model
Meta has open-sourced a massive 800-billion parameter model called Muse Spark Open, which is optimized to run locally on Apple's high-end hardware.
Radwan AbdulhadiEditor and technology writer
Meta Announces Muse Spark Open
Demonstrating its ongoing commitment to open-source software, Meta has introduced a massive new open-weight model called Muse Spark Open. The company has said that it believes the US should not allow Chinese labs to lead on open-source AI development. As part of this effort, Meta is open-sourcing Muse Spark with a model that features an astounding 800 billion parameters.
Meta calls Muse Spark Open "the most capable US open-weight model to date." According to benchmarks released by the company, the model outperforms two leading Chinese models, GLM 5.3 and Kimi K3, specifically on agentic tasks and visual reasoning.
Hardware Requirements and Apple Silicon Integration
Despite its impressive capabilities, running the model locally comes with massive hardware demands. While Muse Spark Open was specifically optimized for Apple Silicon, users will need a staggering 1TB of RAM to run it locally. This makes it the first model optimized to run locally on Apple's powerful machines.
The integration is made possible thanks to work with Apple and key executives. Meta explains that Apple's unified memory architecture lets the M5 Ultra's integrated GPU use up to 512 GB per machine, meaning two systems can hold the entire model. "The result is two Mac Studios behave like one supercomputer with a terabyte of unified memory," Meta notes.
Setup Costs and Configuration Options
Setting up the infrastructure to run Muse Spark Open requires a substantial financial investment. While Apple has not yet announced specific pricing or a definitive ship date for the 512GB configuration of the M5 Ultra-equipped Mac Studio—stating only that it will arrive in late October—any viable setup will be costly.
Alternatively, Muse Spark Open can also run on a configuration consisting of four 256GB machines, which will cost roughly $38,000. While this level of compute sits far out of reach for individual developers or typical hobbyists, larger organizations that regularly burn through expensive token budgets may find the investment worthwhile, provided the model delivers on Meta's performance claims.
Availability and Access
For organizations and developers with the requisite hardware infrastructure, getting started is straightforward. Muse Spark Open weights are officially available to download now from Hugging Face, allowing teams to test the model's performance on agentic tasks and visual reasoning directly.
Sources
- EngadgetMeta built a huge open-weight AI model for Macs, but you'll need 1TB of RAM to run it
Continue chronologically
Related entity coverage
Related stories
Meta Launches Muse Spark Open AI Model for Macs
Meta has released Muse Spark Open, a massive open-weight AI model designed to run locally on Apple silicon using Mac Studio computers.
Meta and Apple Team Up for Muse AI Launch on Mac
Meta and Apple collaborate on the new Muse Spark Open AI model, enabling efficient execution on Mac hardware using the latest MLX framework.
Meta Releases Muse Spark Open Multimodal MoE Model
Meta has introduced Muse Spark Open, an open-weight frontier multimodal mixture-of-experts model featuring 816 billion parameters.
Apple Tightens macOS Full Disk Access Over AI Risks
Apple is updating macOS Full Disk Access controls, warning that increasingly capable AI agents create new security risks for user data.
