Meta releases Muse Spark Open AI model requiring 1TB of Mac memory

Meta released Muse Spark Open on October 15, an 816-billion-parameter open-weight model with 42 billion active parameters, a 1-million-token context window, and native multimodal support for text, images, video, audio, and PDFs, as reported by Engadget and startupfortune.com. Behind that raw power lies a hardware requirement that prices out almost any individual user: running the model demands roughly 1TB of unified RAM, a threshold that forces developers to link multiple high-end Mac Studio machines together over Thunderbolt.

That hardware barrier turns a free, open-weights download into a five-figure infrastructure investment.

Old rivals find common ground under John Ternus

The collaboration bridges a historic chasm between Meta and Apple, two companies that spent years in open hostility. The relationship hit rock bottom after Apple introduced privacy changes in 2021 that limited ad targeting, a move Meta estimated would cut its 2022 revenue by $10 billion, according to CNBC. Public friction followed, with Tim Cook criticizing Meta during the 2018 Cambridge Analytica scandal and Mark Zuckerberg telling Joe Rogan last year that Apple hasn't invented anything great since the iPhone, as noted by CNBC.

Those past disputes contrast sharply with Thursday's announcement, which follows months of cooperation. CNBC reported that representatives from both companies declined to say whether the Muse Spark Open partnership marks the start of a deeper relationship. The shift coincides with leadership changes at Apple, where John Ternus succeeded Tim Cook as CEO last month. Zuckerberg celebrated the release on X, writing, Muse Spark Open -- weights releasing today -- runs great on new Mac Studios thanks to work with @johnternus and Apple, according to CNBC. Ternus responded on X, saying, Congrats to the team! Amazing to run a model of Muse Spark Open's caliber and scale this efficiently, right from your desk. Mac Studio keeps pushing the boundaries of on-device AI! CNBC noted.

What hardware does a terabyte of memory require?

Apple's unified memory architecture is the reason this is even possible at all, allowing the CPU, GPU, and Neural Engine to draw from a shared memory pool rather than a capped VRAM setup, startupfortune.com explained. The top configuration of the new Mac Studio, built around Apple's M5 Ultra chip, tops out at 512GB of unified memory. Linking two maxed-out machines over Thunderbolt creates a single terabyte pool capable of holding the model, according to startupfortune.com and Engadget. Alternatively, developers can configure four 256GB M5 Ultra Mac Studios to reach the same 1TB capacity.

Apple has not yet announced specific pricing or a ship date for the 512GB Mac Studio beyond a late October launch window, Engadget reported. Industry pricing chatter places the 512GB units well north of $10,000 each, bringing a two-machine pair to roughly $20,000 to $25,000, startupfortune.com noted. A four-machine setup featuring 256GB units will cost approximately $38,000, according to Engadget. Meta AI chief Alexandr Wang acknowledged in an interview covered by CNBC that many powerful open-weight models are too big to be run on the majority of personal computers, but argued that Apple's hardware makes enough memory available to users so that it's practical to run Muse Spark Open.

We worked with Apple to bring leading Agent capabilities to your desktop for the first time on the Apple M5 Ultra Mac Studios, so you know we're really excited about this.

Alexandr Wang, Meta AI chief, via CNBC

That amount of compute is pretty squarely out of reach for the typical vibe coder, though larger organizations that've been burning through their token budgets may find investing in a few Mac Studios a pretty good deal if performance matches Meta's claims, Engadget observed. Developers can download a version of the AI model from websites like Hugging Face without extra configuration for Apple's silicon, according to CNBC and Engadget.

How does the open-weight landscape compare right now?

Meta calls Muse Spark Open the most capable US open-weight model to date, stating that company benchmarks show it outperforming two leading Chinese models, GLM 5.3 and Kimi K3, on agentic tasks and visual reasoning, as reported by Engadget. The release arrives amid an aggressive open-weight arms race. Mistral AI unveiled a 1.05-trillion-parameter model nicknamed Le Chonk on October 6, featuring 49 billion active parameters and a 1-million-token context window, with weights scheduled for release on October 27, startupfortune.com reported. Nvidia took a different path with Nemotron 3.5 Lightning, a 30-billion-parameter mixture-of-experts model with only 3 billion active parameters at a time, released under a license that permits commercial use, according to startupfortune.com.

While Meta positions Muse Spark Open as a model that a lot of companies and developers will use and tinker with at small scale before seeking to ramp up and scale up their usage, Wang noted that the company has also released more advanced proprietary models such as Muse Wonder earlier in the week, alongside a smaller open-weight model called Muse Glimmer in August, according to CNBC. For almost everyone outside an enterprise cluster, the honest answer remains Meta's own cloud API, or nothing, startupfortune.com concluded.

The hardware math is further complicated by broad memory market inflation. Goldman Sachs projected a DRAM undersupply of roughly 4.9% this year, calling it the most severe memory shortage in over 15 years as data centers buy up HBM capacity and squeeze wafer supply, startupfortune.com noted. That leaves developers facing steep hardware costs at the exact moment component prices are climbing.

Meta released Muse Spark Open on October 15, 2026.