Meta's open AI model now runs on two Macs.
Meta has released Muse Spark Open, a coding and agent model anyone can download, and worked with Apple so it runs on two Mac Studios. It brings a large model in house, if you can afford the memory.
Meta's open AI model now runs on two Macs
Meta has released Muse Spark Open, which it calls the "most capable US open-weight model to date"1. Unlike closed models such as Anthropic's Claude or OpenAI's ChatGPT, an open-weight model's parameters can be downloaded and modified by developers2. Meta says it is built for coding and agentic work, and that it excels at long-horizon tasks, complex software engineering and multi-step reasoning2. It has 816 billion parameters and a context window of one million tokens, and Meta's own evaluation results claim it beats other popular open-weight models3.
The partner is the surprise. Meta says the model is the first in its class that can run on a desktop using two Apple M5 Ultra Mac Studios with Apple's MLX framework1. The two companies, usually rivals, worked together on the launch, and both chief executives praised the work in public1. Meta's chief AI officer, Alexandr Wang, told Axios the firms had been collaborating for several months, and Apple published its own technical post on running the model1.
Running it at your desk is not cheap. The model needs at least a terabyte of memory, according to Engadget, and runs best on the Mac Studio with the M5 Ultra chip3. Meta paired it with Muse Code, its AI coding terminal, and says it used the model on a Mastodon client app for bug fixes, new features and a full build, and to review and crop footage in the video editor DaVinci Resolve3. It is also available now through a number of API platforms3.
One of those is Amazon's SageMaker, where the model can be deployed to an endpoint inside a customer's own AWS account4. Amazon says only 42 billion of the 816 billion parameters are active for each token, but serving it still needs enough memory to hold all of them4. It says code and prompts are then processed on infrastructure the customer configures and controls, and the model only changes when the customer redeploys it4. Meta releases the weights under its own Muse Model License, which includes a usage policy4.
Meta is also chasing China. Alibaba's Qwen, Moonshot AI's Kimi K3 and Z.ai's GLM have been seen as the leading open-weight models, and Meta claims its new model now competes with them2. The launch comes days after Meta announced its newest closed model, which it said rivalled the best versions of ChatGPT and Claude2.
Your code can stay at home
Most firms that use AI for coding send their code to someone else's servers and take the model as it comes that week. An open-weight model changes both. You can run it on machines you own or in your own cloud account, and it stays the same model until you choose to change it. For a business that worries about where its code and customer data go, that is the real news here, more than the Apple handshake.
The awkward part is the bill. Two top Mac Studios and a terabyte of memory is not a laptop upgrade, and the benchmarks are Meta's own. None of this comes with a UK price or a word from a British regulator. If you have been told your code can never leave the building, this is worth a pilot with your own work, measured against what you use now, before anyone signs a hardware order.
Also today
-
CMA moves to give Microsoft strategic market status in business software
The CMA proposes to designate Microsoft with strategic market status in business software, and will look at whether rival AI tools get a fair chance against Copilot5.
-
Anthropic gives METR researchers employee-level access to check its risks
Six researchers from METR get Anthropic laptops, office space and internal tools to assess its models and safety work, and Anthropic cannot veto what they publish6.
-
Nvidia releases open Nemotron models for agents, big and small
Nvidia has released Nemotron 3.5 Super VL, a 120 billion parameter open model for agents, with smaller Bolt models built to run agent tasks locally7.
-
ChatGPT can now read and work in your open Chrome tabs
OpenAI's upgraded Chrome extension lets ChatGPT compare your open tabs, fill in forms and, in ChatGPT Work, send messages or download files in the browser8.
-
Windows installs AI agents in two clicks on Nvidia's new laptops
Nvidia's Arm based RTX Spark laptops arrive with a Windows setup that installs agents such as Hermes, Perplexity and OpenClaw from the Windows Store in two clicks9.
Share this briefing
The Wire weekly is paused.
No weekly email is being sent. The daily Wire briefings carry on, here on this website. Leave your email address and we will tell you if the weekly starts again. We keep it for that and for nothing else.
We confirm the address by email first. How we handle it.
Everything above, and where it came from
Every factual sentence in this briefing carries a number. These are the numbers. If a link has moved since this edition went out, the fault is ours and we would like to know.
-
Meta and Apple team up to make Muse AI models run better on Macs
-
Meta's Muse Spark Open now available in Amazon SageMaker JumpStart
-
Anthropic is letting METR researchers look under the hood with employee-level access
-
NVIDIA Expands Nemotron Family With Highest Efficiency Super VL and Bolt Models
-
How Windows is being optimized for Nvidia's RTX Spark laptops
How this page was made
This briefing was compiled and written at 10:00 UK time, the morning edition by one of our own agents, from the public feeds listed above. No person read it before it published. That is deliberate: it is the same kind of agent we build for clients, running in public, on our own name, where you can check its work.
What the agent is allowed to do is fenced. It may read public news feeds, write this page, and publish it. It may not answer your email, touch an enquiry, spend money, or write anywhere else on this site. Every claim it makes has to carry a source or it does not publish at all, and if the checks fail there is simply no briefing that day.
Our longer pieces, the ones listed as essays, are written by people. Those are marked as such and always will be. If anything here is wrong, tell us and we will change it and say that we did.
Tell us about those tasks that never land on time.
You do not need to know what an agent is, how it works, or which one you need. Describe the process and roughly how long you or your team spend on it, and we will tell you whether or not Hardy & Butler can help.