Meta has introduced Muse Spark Open, a new artificial intelligence model that reportedly ranks as the highest-performing U.S. open-weight model on the Artificial Analysis Intelligence Index, surpassing NVIDIA’s Nemotron 3.5 Super VL.
The launch strengthens Meta’s position in the competitive market for advanced AI models, where developers are increasingly focused on performance, computing efficiency, pricing and accessibility.
According to technical specifications released by Meta Superintelligence Labs on October 15, Muse Spark Open features 816 billion total parameters, approximately 42 billion active parameters per token, and a context window of one million tokens.
The model’s release also comes with a custom licensing agreement and a reported collaboration with Apple to support deployment on Mac Studio hardware.
Meta Takes Lead in U.S. Open-Weight AI Ranking
Muse Spark Open reportedly overtook NVIDIA’s Nemotron 3.5 Super VL on the Artificial Analysis Intelligence Index, a benchmark used to compare the capabilities of leading AI models.
Open-weight models allow developers and organizations to access model parameters, potentially enabling deployment and customization outside the original developer’s hosted infrastructure.
However, open-weight availability does not necessarily mean a model is fully open source. Licensing agreements can impose restrictions on how the technology is distributed and used.
Meta’s model uses a mixture-of-experts architecture, in which only a portion of the model’s total parameters is activated for a given task.
Although Muse Spark Open has 816 billion parameters overall, approximately 42 billion are active during processing.
This design is intended to provide substantial model capacity without requiring every parameter to be used for each request.
Pricing Highlights Differences Across AI Providers
According to pricing published by OpenRouter, Meta’s Muse Spark Open costs $1.25 per million input tokens and $4.25 per million output tokens through the platform.
That is considerably higher than NVIDIA’s Nemotron 3.5 Super VL, which is listed at $0.10 per million input tokens and $0.40 per million output tokens.
DeepSeek V4.1 Flash is listed at $0.006 per million input tokens and $1.20 per million output tokens at its peak rate.
Google’s Gemini 3.8 Flash, meanwhile, is priced at $0.75 per million input tokens and $3.75 per million output tokens under rates reportedly available through December 31.
The differences illustrate how AI providers are positioning their models across performance and pricing categories.
However, token pricing alone does not establish the total cost of using a model. Computing requirements, deployment arrangements and workload characteristics can also influence overall expenses.
Meta Introduces Custom Licensing Requirements
Muse Spark Open is distributed under Meta’s custom Muse Model license, which requires users to provide attribution using the phrase “Built with Muse Spark Open.”
The requirement distinguishes the model from some alternatives distributed under more permissive open-source licenses.
Licensing terms have become an important consideration in the AI industry as companies release increasingly capable models while maintaining different levels of control over their commercial use and distribution.
Meta and Apple Collaborate on Mac Studio Deployment
According to an official announcement published by Meta, the company collaborated with Apple to optimize Muse Spark Open for Mac Studio computers powered by the M5 Ultra chip, allowing developers to run the AI model locally using Apple’s MLX framework.
The collaboration could expand the model’s availability beyond conventional cloud-based infrastructure by supporting deployment on Apple’s professional computing hardware.
The reported arrangement illustrates growing interest in running advanced AI models across different computing environments, including local systems and specialized hardware.
However, the supplied information does not specify the hardware configurations, memory requirements or performance levels expected for the Mac Studio deployment.
