Meta has recently unveiled its latest artificial intelligence model, Muse Spark Open—a multimodal AI model featuring 816 billion parameters and the capacity to process up to 1 million tokens of input and output context simultaneously.
Muse Spark Open offers multimodal support (the ability to process data from various sources, including text, audio, video, etc.) and is specifically designed to enable developers not only to build diverse artificial intelligence technologies but also to deploy the model as an AI agent for a wide range of tasks.
However, Meta notes that for such applications, companies or developers will require a computer with at least 1TB of memory to utilize Muse Spark Open optimally.
A unique aspect of the Muse Spark Open AI model is that it can be used optimally when deployed on systems like the Mac Studio, powered by Apple’s latest M5 Ultra processor. Meta states that with this hardware (configured with 256GB of memory), users can leverage the AI model by connecting two Mac Studio units in a daisy-chain configuration.
Alternatively, users have the option to access Muse Spark Open via the cloud through API platforms such as Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter, and Together AI. The open-source model—or the AI model weights—can also be downloaded from digital repositories including Amazon SageMaker AI, Google Cloud Model Garden, Hugging Face, and Microsoft Foundry Models.
