{"slug": "can-you-run-muse-spark-open-locally-ai-hype-vs-reality", "title": "Can You Run Muse Spark Open Locally? AI Hype vs Reality", "summary": "Meta released Muse Spark Open on Thursday, an 816-billion-parameter open-weight model with 42 billion active parameters per token and a 1-million-token context window, which requires about 493 GB of memory at 4-bit quantization, more than 1.6 TB at full BF16 precision, and 180 to 212 GB for its smallest 1-bit version. Meta's own developer guide states the model runs locally on Macs only by pairing 2 Mac Studios with 512 GB of memory each or 4 with 256 GB each, and Apple's MLX team wrote that \"the model cannot fit on a single machine\"; VentureBeat noted Apple's 512 GB configuration does not ship until late October. Simon Willison wrote on the launch thread that he \"had to run them via OpenRouter - I don't have access to a machine that can handle this size of model,\" and Meta's smaller Muse Glimmer still needs at least 26 GB of RAM per LM Studio.", "body_md": "# Can You Run Muse Spark Open Locally? AI Hype vs Reality\n\n**Only on a computer with hundreds of gigabytes of memory. Muse Spark Open needs about 493 GB in the 4-bit format Apple ran it in, and even its smallest 1-bit version needs 180 GB. My MacBook Pro has 16.** [Meta](https://www.meta.com/) released it on Thursday as “[the most capable US open-weight model to date](https://research.meta.ai/blog/muse-spark-open)” and “the first model in its class to run on your desktop”. This post is about that desktop and the *US* in that sentence. And it’s about why I think AI launches like this one keep promising more than they deliver.\n\n## Can you run Muse Spark Open locally?\n\nOn Thursday I asked [Claude Code](https://claude.com/claude-code) in my terminal the obvious question: “can i run muse spark open on this mac”. The answer was a pretty firm no!! This Mac is a 2021 [MacBook Pro](https://www.apple.com/macbook-pro/) with an M1 Pro chip and 16 GB of memory.\n\n**Muse Spark Open runs locally on Macs only if you join 2 [Mac Studios](https://www.apple.com/mac-studio/) with 512 GB of memory each, or 4 with 256 GB each.** That’s from Meta’s own [guide for developers](https://dev.meta.ai/resources/blog/build-with-muse-spark-open). Apple’s machine learning team [wrote up how they did it](https://machinelearning.apple.com/research/mlx-distributed-inference) with [MLX](https://github.com/ml-explore/mlx) (Apple’s open source framework for running models on its own chips). They’re refreshingly blunt about it: “the model cannot fit on a single machine.”\n\nAnd the 512 GB Mac Studio isn’t even on sale yet!! [VentureBeat](https://venturebeat.com/technology/meta-unveils-muse-spark-open-the-new-top-u-s-open-weights-model-and-it-runs-on-two-paired-mac-desktops) checked: “Apple’s 512GB-memory configuration doesn’t arrive until late October.” So the desktop in Meta’s headline is one nobody can buy, at a price nobody can tell you. What you *can* buy today is 4 of the 256 GB models:\n\nA token is roughly a word and that’s faster than anyone can read. It’s also more than twice the price of the [$18,000 AI server](https://tej.as/blog/local-llm-vs-cloud) everyone on X was arguing about this week. I worked out that one needs about 9 months to cost less than 10 seats of a $200 cloud plan. And it’s the *cheap* one.\n\n[Simon Willison](https://simonwillison.net/), who tries new models the day they come out, couldn’t run it either. On the launch thread he [wrote](https://news.ycombinator.com/item?id=50108941) “I had to run them via OpenRouter - I don’t have access to a machine that can handle this size of model.”\n\nMeta’s small model, [Muse Glimmer](https://lmstudio.ai/models/muse-glimmer), is the one meant for computers like mine. [LM Studio](https://lmstudio.ai/) says even that one needs “at least 26 GB of RAM”. So no luck there either!\n\n## How much RAM do you need to run Muse Spark Open?\n\n**Muse Spark Open needs about 493 GB of memory at 4-bit, more than 1.6 TB at full precision and 180 to 212 GB for the smallest 1-bit version.** It’s big because it’s built big: an open-weight large language model (LLM) with 816 billion parameters, 42 billion of them active for each token, and a context window of 1 million tokens. In a [mixture of experts](https://huggingface.co/blog/moe) (MoE) model like this one, every layer holds 256 small expert networks and picks 8 of them for each token ([Hugging Face’s write-up](https://huggingface.co/blog/meta-muse-spark-open) has the details). That’s how a model this big still answers at a usable speed: most of it sits idle on any given word. Idle experts still take up memory though (all 256 of them, in every layer).\n\nFull precision is BF16 (brain floating point, the 16-bit format the model was trained in), where Apple says the weights alone take “more than 1.6TB of RAM”. [Quantization](https://huggingface.co/docs/optimum/concept_guides/quantization) shrinks that by storing each number in fewer bits. In NVIDIA’s 4-bit floating point format (NVFP4) the model takes “about 493GB” and Apple’s team still had to split it across 2 machines because of the memory it needs while it’s working. The 1-bit version comes from [Unsloth](https://unsloth.ai/docs/models/muse-spark-open), who squeeze models down so they fit on smaller machines. It’s still more than 11× my laptop.\n\n## Is Muse Spark Open really open source?\n\n**Muse Spark Open is open weights and not open source. You can download its weights for free under Meta’s own [Muse Model License](https://huggingface.co/meta-models/Muse-Spark-Open-816B-A42B).** To download it you give Meta your “full legal name, date of birth, and full organization name”. Anything you build with it has to say “Built with Muse Spark Open”. A company making more than $20 million a year that offers it as a service has to “notify Meta and enter into a separate written agreement” (all of it is on the [model card](https://huggingface.co/meta-models/Muse-Spark-Open-816B-A42B)). [Open weights](https://tej.as/blog/what-are-open-weights) means you get the finished model to run yourself (the data and the code that made it stay with Meta).\n\nMeta can ship an open source license when it wants to. Muse Glimmer, the small one, is under [Apache 2.0](https://www.apache.org/licenses/LICENSE-2.0). And Muse Spark Open is still more open than the model it’s named after: in 2024 [Mark Zuckerberg](https://en.wikipedia.org/wiki/Mark_Zuckerberg) wrote “[Meta is committed to open source AI](https://about.fb.com/news/2024/07/open-source-ai-is-the-path-forward/).” In April this year Muse Spark became [its first frontier model without open weights](https://the-decoder.com/metas-muse-spark-is-its-first-frontier-model-and-its-first-without-open-weights/).\n\n## Is Muse Spark Open really that good?\n\n**Muse Spark Open is very good for an open-weight model. Meta’s claim only covers open-weight models from American companies (a much smaller field).** [Artificial Analysis](https://artificialanalysis.ai/) ranks it #8 of 117 open-weight models in its size class, with 39 on its [Intelligence Index](https://artificialanalysis.ai/models/muse-spark-open) (the median for that class is 19 so it’s good!). Meta’s own model card puts it behind [Kimi K3](https://huggingface.co/moonshotai) from the Chinese lab [Moonshot AI](https://www.moonshot.ai/) on both of the coding benchmarks it reports:\n\nVentureBeat looked at the same table and called it “a contender with specific strengths, rather than an overall open-weight leader.” That’s fair! It’s just not what “the most capable US open-weight model” sounds like at a glance. [Alexandr Wang](https://x.com/alexandr_wang), Meta’s chief AI officer, even ended his [launch post](https://x.com/alexandr_wang/status/2110778736434184611) with “america! 🇺🇲”.\n\nMeta has been here before. When it launched [Llama 4](https://ai.meta.com/blog/llama-4-multimodal-intelligence/) in April 2025, the announcement led with “an experimental chat version scoring ELO of 1417 on [LMArena](https://lmarena.ai/)”. LMArena [said afterwards](https://x.com/lmarena_ai/status/1909397817434816562) that “Meta should have made it clearer that ‘Llama-4-Maverick-03-26-Experimental’ was a customized model to optimize for human preference.” The biggest Llama 4, Behemoth, was “still training” at launch. A month later [Axios reported](https://www.axios.com/2025/05/15/meta-behemoth-llama-scaling-delays), citing the [Wall Street Journal](https://www.wsj.com/), that Meta had delayed it over “concerns that it may not be enough of an advance.”\n\n## Is AI overhyped?\n\n**A lot of AI launches are. The tools work, but launch posts keep promising more than the product does on the day you get it.** I said this on stage at [React fwdays](https://fwdays.com/en/) in 2024, about my then brand new iPhone 16 Pro Max that still didn’t have [Apple Intelligence](https://www.apple.com/apple-intelligence/) in Europe:\n\nthere’s devices like the R1, the Rabbit R1 that’s claiming to do a bunch of things and then I get it and it doesn’t do that. I even feel that with Apple. I was promised Apple Intelligence…\n\nA few seconds later I said “there’s a lot of overpromise, underdeliver and just AI AI AI everywhere and it’s really exhausting.” Here’s that bit of the talk:\n\nApple [said in March 2025](https://daringfireball.net/2025/03/apple_is_delaying_the_more_personalized_siri_apple_intelligence_features) that “It’s going to take us longer than we thought to deliver on these features”. When I finally used Apple’s on-device model in my own app this year, it [refused a voice note](https://tej.as/blog/note-app-ai-benchmark-apple-intelligence-sucks) of a normal conversation with my wife as “unsafe”. [Humane](https://en.wikipedia.org/wiki/Humane_Inc.) [raised more than $230 million](https://techcrunch.com/2025/02/18/humanes-ai-pin-is-dead-as-hp-buys-startups-assets-for-116m/) for its AI Pin, sold its assets to [HP](https://www.hp.com/) for $116 million and switched every Pin off on February 28, 2025. A couple of weeks ago the US government started [calling AI “Super Intelligence”](https://tej.as/blog/super-intelligence-isnt-superior) without changing its definition.\n\n[Marques Brownlee](https://www.youtube.com/@mkbhd) reviewed the Rabbit R1 and the AI Pin in 2024. In his [Rabbit R1 review](https://www.youtube.com/watch?v=ddTV12hErTc&t=928s) he called AI products “the apex of this horrible trend where the thing that you get at the beginning is like borderline nonfunctional compared to all the promises”. His advice near the end is the best test I know for a launch like Thursday’s: “[buy the product based on what it is today and not what it’s promised to be in the future](https://www.youtube.com/watch?v=ddTV12hErTc&t=1155s).”\n\nApplied to Muse Spark Open, his test says it’s a very good open model that runs on hardware almost nobody has today (and through an API for everyone else). That’s a fine thing to ship! It’s just a different thing from a model that runs on your desktop.\n\nPeople believe the promise even after it doesn’t land. In a [2025 study](https://metr.org/blog/2025-07-10-early-2025-ai-experienced-os-dev-study/) by [Model Evaluation and Threat Research](https://metr.org/) (METR), experienced open source developers expected AI to speed them up by 24%. They took 19% longer with it. Afterwards they “still believed AI had sped them up by 20%”!! METR has since marked those results “out of date” and [its 2026 update](https://metr.org/blog/2026-02-24-uplift-update/) says developers are likely faster with AI tools now. Good! I still keep thinking about the gap between how fast it felt and how fast it was.\n\n## AI should be like salt\n\nRight after the Rabbit R1 bit in that talk I said “AI features should be like salt”. You “sprinkle them on your products and you can create true beauty and flavor”, but if you drown a product in it “people get sick”. Muse Spark Open is a lot of good engineering, from 256 experts in every layer to the MLX work that makes 4 Macs behave like one computer. I just wish Meta had launched it the way Apple’s engineers wrote it up, as a great model that doesn’t fit on one machine.\n\n## Questions\n\n### Can I run Muse Spark Open locally?\n\nOnly on a machine with hundreds of gigabytes of memory. Muse Spark Open needs about 493 GB in the 4-bit format Apple ran it in and 180 GB for the smallest 1-bit version. So Meta's own guide asks for 2 Mac Studios with 512 GB each or 4 with 256 GB each. My 16 GB MacBook Pro can't run it, which leaves me with an API.\n\n### Will Muse Spark Open work on a Mac?\n\nYes, if you join 2 Mac Studios with an M5 Ultra and 512 GB of memory each, or 4 with 256 GB each, using Apple's open source machine learning framework. Apple's own machine learning team says the model can't fit on a single machine. The 512 GB Mac Studio doesn't go on sale until late October 2026 and 4 of the 256 GB models cost $43,196 before tax.\n\n### How much RAM do you need to run Muse Spark Open?\n\nAbout 493 GB in NVIDIA's 4-bit format, more than 1.6 TB at full 16-bit precision, and 180 to 212 GB for Unsloth's smallest 1-bit version. Apple adds that the memory the model needs while it works means even the 4-bit version can't fit on one machine. Meta's smaller Muse Glimmer model needs at least 26 GB of RAM.\n\n### Is Muse Spark Open open source?\n\nNo. Muse Spark Open is open weights and not open source: its weights are free to download under Meta's own Muse Model License. To download it you give Meta your full legal name and date of birth. A company making more than $20 million a year that offers it as a service needs a separate agreement with Meta.\n\n### Is Muse Spark Open really that good?\n\nIt's very good for an open-weight model. Artificial Analysis ranks it #8 of 117 open-weight models in its size class, with 39 on its Intelligence Index against a median of 19. Meta's own model card puts it behind Kimi K3 from Moonshot AI on both coding benchmarks it reports: 76.4 against 85.0 on Terminal-Bench 2.1 and 60 against 68 on DeepSWE 1.1.\n\n### Is AI overhyped?\n\nMany AI launches are. Meta announced Muse Spark Open as \"the first model in its class to run on your desktop\". That desktop is 2 Mac Studios with a memory option Apple doesn't sell until late October 2026. Marques Brownlee's advice from his Rabbit R1 review is the test I'd apply: buy the product based on what it is today.\n\nWritten by me, Tejas Kumar, an AI Engineer at IBM based in Berlin. Read [everything else I have written](https://tej.as/blog), or go to [Fluent React, my O'Reilly book on how React works inside](https://tej.as/react), [the talks I give at conferences](https://tej.as/speaking), and [ConTejas Code, my podcast](https://tej.as/podcast).", "url": "https://wpnews.pro/news/can-you-run-muse-spark-open-locally-ai-hype-vs-reality", "canonical_source": "https://tej.as/blog/can-you-run-muse-spark-open-locally/", "published_at": "2026-10-16 12:00:00+00:00", "updated_at": "2026-10-16 14:49:57.612417+00:00", "lang": "en", "topics": ["large-language-models", "ai-infrastructure", "ai-products", "artificial-intelligence"], "entities": ["Meta", "Muse Spark Open", "Muse Glimmer", "Apple", "MLX", "Mac Studio", "Simon Willison", "LM Studio"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/can-you-run-muse-spark-open-locally-ai-hype-vs-reality", "markdown": "https://wpnews.pro/news/can-you-run-muse-spark-open-locally-ai-hype-vs-reality.md", "text": "https://wpnews.pro/news/can-you-run-muse-spark-open-locally-ai-hype-vs-reality.txt", "jsonld": "https://wpnews.pro/news/can-you-run-muse-spark-open-locally-ai-hype-vs-reality.jsonld"}}