Meta releases Muse Spark Open, an 816-billion-parameter open-weight AI model built to run on Macs
Meta calls it the most capable open-weight model in the US and says it runs best on a Mac Studio with Apple's M5 Ultra. The catch: it needs about 1TB of memory.

Meta has released Muse Spark Open, a free-to-download open-weight AI model with 816 billion parameters and a 1 million token context window, TechCrunch reported on Thursday. Meta calls it the most capable open-weight model in the United States, and it has worked with Apple so the model runs locally on Macs.
An open-weight model is one whose trained weights anyone can download and run on their own hardware, instead of only reaching it through the company's servers. That matters for developers, universities and companies in Pakistan that want to use a large model without sending their data abroad or paying per message in dollars.
What Meta announced
- Size: 816 billion parameters, with a context window of 1 million tokens, enough to read a very large codebase or several long documents at once.
- Abilities: natively multimodal, and built for complex coding and agent tasks, according to Meta.
- Hardware: it runs best on a Mac Studio with the M5 Ultra chip. Engadget, cited by TechCrunch, reported that it needs at least 1TB of memory.
- Access: available now through several API platforms for those without that kind of machine.
- Benchmarks: Meta published evaluation results on Thursday and says the model beats other popular open-weight models. TechCrunch did not list the scores.
In demonstrations, Meta paired the model with Muse Code, its AI coding terminal, to fix bugs, add features and build a Mastodon app, and used it with DaVinci Resolve to review and crop video footage.
Who can actually run it
Very few people. A machine with 1TB of memory is far beyond an ordinary laptop or desktop, so most users in Pakistan will meet Muse Spark Open through an API or a hosted service rather than on their own computer. For software houses and research labs, the open weights mean they can, with enough hardware, host it themselves.
This is Meta's second Muse release in a week. On 12 October it launched Muse Wonder as a paid service at $20 or $100 a month (our story on what that costs in rupees).
What it costs you
The model itself is free to download. Hosted API access is billed in dollars, so the rupee cost moves with the exchange rate: follow it on our dollar rate page and convert any plan with the currency converter. Freelancers paying for AI tools out of foreign income can check their tax position with the freelancer tax calculator.
Frequently asked questions
- What is Muse Spark Open?
- An open-weight AI model from Meta with 816 billion parameters and a 1 million token context window, released on 15 October 2026 and designed to run on Macs.
- Can I run Muse Spark Open on my laptop?
- Almost certainly not: it reportedly needs at least 1TB of memory and runs best on a Mac Studio with Apple's M5 Ultra. Most people will use it through an API.
Sources
More in AI
AI
FTC subpoenas Anthropic and OpenAI over AI safety incidents
Axios reports each subpoena exceeds 50 pages and seeks records on security incidents and safeguards. The FTC probe is investigative, with no finding of wrongdoing.
12h ago
AI
FTC subpoenas Anthropic and OpenAI over AI safety incidents
Axios reports each subpoena exceeds 50 pages and seeks records on security incidents and safeguards. The FTC probe is investigative, with no finding of wrongdoing.
12h ago
AI
Nvidia pledges $250 million in computing to UK AI safety research over five years
Jensen Huang's $50 million a year in compute will go to agentic safety testing at Oxford, Imperial and UCL.
1d ago



Reader comments 0
No comments yet. Say something useful.