Meta is bringing one of its largest open-weight AI models to developers who want more control over where and how their AI workloads run.
Announced Oct. 15, Muse Spark Open brings 816 billion total parameters, 42 billion active parameters per token via a mixture-of-experts design, and a 1-million-token context window. Meta declared the release “the most capable US open-weight model to date.”
Meta calls it the most capable US open-weight model to date, but benchmark comparisons suggest that Chinese competitors remain ahead in some areas. For developers, the bigger question is whether the model’s capabilities justify the hardware and deployment costs.
Bridging silicon and desktop compute
Meta engineered the system alongside Apple’s MLX framework so that two Mac Studio desktops equipped with M5 Ultra chips operate together over a unified memory pool.
“The result is two Mac Studios behave like one supercomputer with a terabyte of unified memory,” Meta stated.
During internal testing, the company used the model inside Muse Code to ship code changes to Ice Cubes, an open-source Mastodon client, spanning “bug fixes, new features, [and] full build.” Meta also showed the model controlling DaVinci Resolve to review video footage frame-by-frame and crop vertical clips.
The release is available immediately as downloadable weights across Hugging Face, Amazon SageMaker AI, Google Cloud Model Garden, and Microsoft Foundry Models. Cloud API hosting is also available via platforms like Fireworks AI, Together AI, and Groq.
Benchmark realities behind the domestic badge
While Meta highlighted strong gains against domestic open systems like Inkling and Nemotron 3 Ultra, global open benchmarks reveal a fragmented landscape.
Meta’s internal evaluation showed Spark Open leading rivals on tool-calling evaluations, scoring 89.2 on MCP-Atlas and 61.8 on JobBench. Yet Chinese open-weight models retained the upper hand on pure coding benchmarks: Moonshot AI’s Kimi K3 hit 85 on Terminal-Bench 2.1, outpacing Spark Open’s 76.4.
Independent ratings from Artificial Analysis placed Spark Open at an Intelligence Index score of 39, trailing Chinese models such as Xiaomi’s MiMo-V2.6-Pro at 46, GLM 5.3 at 45, and Kimi K3 at 44.
The results suggest that Meta’s model could be competitive for agentic workflows without necessarily leading the broader open-weight market.
Why a US-developed open-weight model matters
Meta’s emphasis on Muse Spark Open as a US-developed model adds another dimension to the competition with Chinese AI providers.
By delivering a viable American alternative to high-ranking models from Alibaba, Z.ai, and Moonshot AI, Meta is catering to enterprise procurement officers bound by corporate governance or regulatory mandates that restrict the deployment of foreign-developed systems.
Rather than trying to beat Chinese labs on every terminal-coding metric, Meta is packaging sufficient multi-step agent reasoning into an auditable domestic package that satisfies institutional compliance hurdles.
More must-read AI coverage
- SS&C Intralinks DealCentre AI vs. Datasite: Which platform is built for the future of dealmaking?
- SS&C Intralinks FundCentre AI vs. Juniper Square: Which platform better supports modern private markets fund managers?
- Why Data, Not Models, Determines AI Success
- The Rise of the AI-Native Factory: How Physical AI Is Transforming Manufacturing
Implications for software teams and developers
For software developers and IT leaders, Muse Spark Open reshapes deployment choices:
- Hardware cost barriers: Running the model locally requires 1 terabyte of memory. Apple’s 512GB M5 Ultra Mac Studios will not arrive until late October with unannounced pricing, while an equivalent four-machine 256GB array costs over $38,000.
- Licensing limitations: The Muse Model License permits free commercial modification, but requires a custom agreement for model-serving businesses that exceed $20 million in 12-month gross revenue. Products must also display “Built with Muse Spark Open.”
- Procurement screening: Downloading via Hugging Face requires legal names, birth dates, and corporate identifiers, establishing a strict registration hurdle for engineering teams.
For enterprise developers, Muse Spark Open offers another option for running advanced AI workloads outside fully managed cloud services.
Whether local deployment makes financial sense will depend on hardware requirements, licensing terms, performance, and expected usage. Organizations considering the model should evaluate those trade-offs against cloud-hosted alternatives before committing to dedicated infrastructure.
Want to get better results from AI? TechRepublic readers can take The Neuron Academy’s free six-minute How to Talk to AI course to learn simple prompting techniques that make AI more useful at work. You can also enjoy free seven-day access to The Neuron Academy’s other practical AI lessons. Explore all courses →