Meta has released Muse Spark Open, the first open-weight model in its Muse Spark series. The multimodal AI features 816 billion total parameters, with 42 billion active, and supports a 1 million-token context window. It accepts image, video, and audio inputs.
The model is now available for public testing on Arena.ai’s Agent Arena and Code Arena platforms. Meta positions the tool for long-horizon agentic workflows, complex coding tasks, and multi-step reasoning, including repository-wide analysis.
Developers can access the model via API providers such as Baseten and Fireworks AI, or download weights from Amazon SageMaker and Google Cloud Model Garden. The model is optimized for Apple silicon using the MLX framework.
Meta recommends a reference infrastructure consisting of two M5 Ultra Mac Studios with 1TB of unified memory. The company states that the model underwent safety post-training, allowing it to function without external guardrails.
