Recommended Free Tools
iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Meta announced Muse Spark Open on October 15, 2026. The open-weight, natively multimodal model has 816 billion total parameters, 42 billion active parameters and a claimed 1 million-token context window. Meta says it can run locally across two M5 Ultra Mac Studios linked through Apple’s MLX framework; the announcement does not show that one standard Mac Studio can hold the complete unquantized model.
What Meta announced
Muse Spark Open is aimed at long-horizon tasks, complex software engineering and agentic workflows in which an AI system must plan and complete multiple steps. Making the weights available allows organizations to self-host, customize and run the model through supported inference frameworks instead of relying only on a hosted chatbot or API.
| # | Preview | Product | Price | |
|---|---|---|---|---|
| 1 |
|
Apple Studio Display: Standard Glass, Tilt-Adjustable Stand | $1,599.00 | Buy on Amazon |
| 2 |
|
Apple 2026 MacBook Neo 13-inch Laptop with A18 Pro chip: Built for AI and Apple Intelligence, Liquid... | $789.99 | Buy on Amazon |
| Specification | Meta’s launch figure | How to interpret it |
|---|---|---|
| Total parameters | 816 billion | The full size of the released model, as stated by Meta AI Research in 2026. |
| Active parameters | 42 billion | The active-parameter figure Meta reports; it does not reduce the storage needed for the complete unquantized weights. |
| Context window | 1 million tokens | Meta’s stated maximum context; real workloads may be constrained by memory, software and serving configuration. |
| Modalities | Natively multimodal | Designed to work across more than text-only inputs and outputs, subject to the application and deployment stack. |
Meta describes Muse Spark Open as its most capable open-weight model, but no independent benchmark table accompanied the launch. That description should therefore be treated as a vendor claim rather than a measured ranking against other local models.
Can Muse Spark Open run on a Mac Studio?
Yes, for the configuration Meta demonstrated: two Mac Studios equipped with M5 Ultra chips. Meta says MLX optimization lets the two machines expose up to 512 GB of unified memory each and operate together as a system with up to 1 TB of unified memory. The M5 Ultra memory bandwidth is stated as 1.2 TB/s.
#1 Best Overall
- A SIGHT TO BE BOLD — An immersive 27-inch 5K Retina display, 12MP Center Stage camera with Desk View, advanced mics and speakers, and Thunderbolt 5 ports.* Studio Display is the perfect partner for Mac.
- WITH FLYING COLORS — 14.7 million pixels, 600 nits of brightness, support for 1 billion colors, and P3 wide color make everything spring to life with spectacular detail.
- ADVANCED CAMERA AND AUDIO — The 12MP Center Stage camera helps you look sharp in any light and keeps you in frame during video calls. Desk View lets you share your workspace. And a studio-quality three-mic array ensures you come through clearly.
- SIX-SPEAKER SOUND WITH SPATIAL AUDIO — Four force-cancelling woofers produce bold bass, and two high-performance tweeters deliver accurate mids and crisp highs. And Spatial Audio creates a sophisticated, cinematic soundstage in supported music, video, and games.
- SUPERPOWER YOUR STUDIO — Two Thunderbolt 5 ports let you connect high-speed accessories, daisy-chain up to four displays, and even fast-charge Mac laptops. And two USB-C ports are perfect for USB peripherals or charging your iPhone.
| Deployment detail | Advertised value | Qualification |
|---|---|---|
| Mac Studios used | 2, each with M5 Ultra | This is Meta’s demonstration setup, not a universal minimum for every quantized or distributed deployment. |
| Unified memory per Mac Studio | Up to 512 GB | Maximum stated by Meta for the M5 Ultra configuration. |
| Combined memory | Up to 1 TB | Meta says the two systems can hold the entire model and behave like one supercomputer. |
| Memory bandwidth | 1.2 TB/s | Meta’s stated M5 Ultra figure; no independent performance test was provided. |
A single ordinary Mac Studio is not established by the launch announcement as being able to hold the full unquantized Muse Spark Open model. Quantization, sharding and future software changes could alter the practical hardware requirement, but Meta did not publish a single-Mac performance result, power test or street-price analysis.
Why the memory distinction matters
The 816-billion-parameter total describes the complete model, while 42 billion is the active-parameter figure Meta reports. Active parameters should not be read as a guarantee that a 42-billion-parameter computer is sufficient: serving the released weights also requires memory for model storage, runtime state, the long context and the inference framework. The exact requirement depends on precision, quantization and serving configuration, none of which Meta quantified in the announcement.
What Meta demonstrated
Software engineering with Muse Code
Meta showed Muse Code making iOS changes to Ice Cubes, an open-source client for the Mastodon social network. The demonstration is evidence of the workflow Meta chose to showcase, not an independent measure of coding accuracy, reliability or the amount of human review required.
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Clear out junk files and repair common Windows errorsFree Scan →Multimodal video work
In another demonstration, the model controlled DaVinci Resolve, inspected footage frame by frame, selected strong moments and cropped them vertically. This illustrates the intended combination of visual understanding, software control and multi-step planning. It does not establish that every supported video-editing task will work without application-specific integrations or supervision.
How developers can access the model
Hosted API providers
Meta lists API access through Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter and Together AI. Availability, pricing, rate limits, regional coverage and the exact model version are controlled by each provider and were not specified in Meta’s announcement.
Model-weight distribution
Meta lists Amazon SageMaker AI, Google Cloud Model Garden, Hugging Face and Microsoft Foundry Models as distribution channels for the weights. A weight download is different from an API subscription: self-managed users remain responsible for hardware, storage, serving software, security and updates.
Rank #2
- AN AMAZING MAC AT A SURPRISING PRICE — With an incredibly portable and durable aluminum design, up to 16 hours of battery life,* and the A18 Pro chip, MacBook Neo is ready to go wherever school takes you.
- FOUR STUNNING COLORS. ONE DURABLE DESIGN — Choose from four beautiful colors — Silver, Blush, Citrus, or Indigo — each with a color-coordinated keyboard. And MacBook Neo is made with a durable recycled aluminum enclosure that helps it reach 60 percent recycled content by weight — the most ever in any Apple product.*
- FLY THROUGH EVERYDAY ASSIGNMENTS — Whether you’re cramming for finals, using Apple Intelligence* to summarize class notes, creating presentations, or even playing the latest Apple Arcade game,* MacBook Neo delivers the performance and AI capabilities you need to get things done.
- UP TO 16 HOURS OF BATTERY LIFE — MacBook Neo delivers all day battery life, so you can power through from early morning classes to late night study sessions without worrying about plugging in.
- A VIBRANT 13-INCH DISPLAY* — The gorgeous Liquid Retina display on MacBook Neo supports 1 billion colors, so photos and videos pop and text is crisp for easy reading.
Inference and customization compatibility
The announced compatibility list includes SGLang, vLLM, TensorRT-LLM, Hugging Face Transformers, Apple MLX, llama.cpp and Unsloth. Meta also identifies PyTorch for customization. Compatibility means the model is intended to work with these tools; it does not guarantee identical features, throughput or memory use across them.
What to check before choosing a local deployment
- Memory capacity: Confirm whether your chosen precision and quantization fit the available unified or accelerator memory. Do not assume the two-Mac demonstration applies to a single machine.
- Context needs: A 1-million-token window is a headline limit, not a promise that your application can process that much text economically or at the same speed as a short prompt.
- Workflow fit: Test the specific coding, visual-control or agent steps you need. The launch demonstrations are vendor examples, not independent benchmark results.
- Serving stack: Verify that your selected framework supports the model’s multimodal inputs, distributed execution, quantization method and desired hardware.
- Operations: Plan for model files, runtime state, monitoring, access controls and update procedures when self-hosting.
Safety disclosures and unanswered questions
Meta says it evaluated Muse Spark Open under its Meta Superintelligence Scaling Framework, including the most severe risks, and that the model met the thresholds required for release. This is Meta’s own safety assessment. The announcement does not provide an independent audit, detailed test results or a reproducible safety benchmark.
The launch materials also leave several practical questions open: there is no independent performance comparison, single-Mac throughput result, power-consumption measurement or street price. Those omissions matter for anyone deciding between local hardware and a hosted API.
Why the Mac deployment is significant
Running a model of this scale across consumer-oriented desktop systems is a proof of concept for local AI that does not depend entirely on a datacenter. Apple’s large unified-memory configurations let the CPU, GPU and other processing resources address the same memory pool, while MLX provides the software path Meta used for the demonstration.
Axios reported that Meta framed Muse Spark Open as its most powerful U.S.-developed open-weight model and as a proof of concept for powerful local systems. Meta AI chief Alexandr Wang summarized the partnership in a lighthearted line: “We’re excited for mommy and daddy to be playing nice again.” The statement reflects Meta’s positioning, not an independent technical evaluation.
The practical takeaway
Muse Spark Open is a major open-weight release with unusually large parameter and context figures, and Meta has shown it operating on two M5 Ultra Mac Studios. For readers considering a desktop deployment, the reliable conclusion is narrower than “it runs on any Mac Studio”: the advertised local setup is a pair of high-memory M5 Ultra systems, while the requirements and performance of a single Mac, different quantization levels and other hardware remain unestablished.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
