AnswersPublished23 sources
What Meta’s Muse Spark Open Means for AI on Mac
Meta released Muse Spark Open’s weights on October 15, 2026, describing it as its most capable U.S. The model handles text, images, audio and video input and is designed for coding and agentic work.
Published byEdited with GPT-6 LunaImages generated with GPT Image 2
Research answer
Key takeaways
- Meta released Muse Spark Open’s weights on October 15, 2026, describing it as its most capable U.S.
- The model handles text, images, audio and video input and is designed for coding and agentic work.
AI Prompt
openai.comCreate a landscape editorial hero image for this Studio Global article: What did Meta announce on October 15 about Muse Spark Open and its partnership with Apple, including its claim to be the most capable U.S. o. Article summary: On October 15, 2026, Meta released Muse Spark Open’s weights, calling it its “most capable U.S. open-weight model to date,” and announced work with Apple to run it locally on M5 Ultra Mac Studios using MLX. [11][13] The . Topic tags: general, general web, user generated. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, watermarks, charts with fa
Meta released the weights for Muse Spark Open on October 15, 2026, calling it its “most capable U.S. open-weight model to date.” The announcement also highlighted work with Apple to run the model locally on high-memory Mac Studio systems using Apple’s MLX framework. 11
The release combines a very large model with a practical trade-off: developers can access its weights, but running the full model locally requires multiple high-end Macs, and its license and benchmark results need careful reading.
Muse Spark Open’s core specifications
Meta describes Muse Spark Open as a natively multimodal mixture-of-experts model with 816 billion total parameters, of which 42 billion are active. It accepts text, images, audio and video, produces text, and has a 1 million-token context window. 11
17
Meta positions it for coding and agentic tasks—work where a model uses tools or takes multiple steps toward a goal. The company also describes long-horizon tasks, complex software engineering and multistep reasoning as intended uses. 11
53
“Open-weight” means developers can download the model’s learned parameters and run them themselves; it does not automatically mean that every use is free of restrictions. 49
How the local Mac setup works
Meta and Apple worked on a local deployment using MLX, Apple’s machine-learning framework. Meta’s developer documentation describes two configurations: two M5 Ultra Mac Studios with 512GB of unified memory each, or four with 256GB each. Both add up to 1TB of memory and are described as sufficient for the NVFP4-quantized version of the model. 11
31
Apple’s unified-memory design allows the GPU to access that memory, but the 1TB figure is the combined capacity across multiple computers—not the memory in one Mac. 12 The available documentation specifies these configurations; it does not establish that every configuration supports the full context window under every workload.
A report quoted a price of $10,799 for a 256GB M5 Ultra Mac Studio and said a 512GB version was expected later. That figure is not a price for the two-machine configuration with 512GB in each computer, and the available reporting does not establish the price of that setup. 38
What the benchmarks do—and don’t—show
Meta’s published benchmark results, as summarized by Engadget, suggest Muse Spark Open performs better than GLM 5.3 and Kimi K3 on some agentic tasks and visual reasoning. Those are specific comparisons based on company-reported results, not proof that Muse Spark Open is the best model for every task. 21
A separate comparison reports a lower overall Artificial Analysis Intelligence Index score for Muse Spark Open than for GLM 5.3 and Kimi K3. That index and the task-specific results measure different things, so the findings are not necessarily contradictory. The available excerpts do not provide enough detail to make a like-for-like assessment across every benchmark. 19
Downloadable weights still come with license terms
The weights are available to developers, but “open-weight” should not be read as “unrestricted commercial use.” CNBC describes the license as allowing limited commercial use. A separate report on the terms says an API provider with more than $20 million in gross revenue during a 12-month period must make a separate agreement with Meta. Check the license itself before building or selling a product based on the model. 4
43
What local inference means for privacy
Running a model locally can let a developer process prompts without sending them to a hosted model service. That can be useful for applications where control over data flow matters, but it is not a blanket privacy guarantee: connected tools, other services or the application itself may still transmit data. The local setup describes where inference runs, not every part of an app’s data handling. 4
11
The partnership is notable because Apple and Meta have been described as longtime rivals. Meta AI chief Alexandr Wang joked that he was “excited for mommy and daddy to be playing nice again,” according to Axios. 3
48
For developers, the clearest takeaway is that Muse Spark Open offers downloadable weights and a documented route to local inference on Apple hardware. Its large memory footprint, license conditions and mixed benchmark picture matter just as much as Meta’s headline claim about the model’s capabilities.
Studio Global AI
Continue your research
This page includes a source-backed answer you can continue inside Studio Global.
People also ask
What is the short answer to "What Meta’s Muse Spark Open Means for AI on Mac"?
Meta released Muse Spark Open’s weights on October 15, 2026, describing it as its most capable U.S.
What are the key points to validate first?
Meta released Muse Spark Open’s weights on October 15, 2026, describing it as its most capable U.S. The model handles text, images, audio and video input and is designed for coding and agentic work.
Sources
research.meta.aiIntroducing Muse Spark Open, The Most Capable US ...
dev.meta.aiGet started with Meta Model API and Muse Code
openaimaster.comMuse Spark Open: Meta’s 816B Open-Weight AI Model
sea.mashable.comApple and Meta are getting along, and the new Muse Spark Open model brought them together