Meta launched Muse Spark Open, an open-weight coding model with 816 billion total parameters, 42 billion active parameters, and 1-million-token context.

Muse Spark Open is a coding-optimized mixture-of-experts (MoE) model with 816 billion total parameters, 42 billion active parameters, and a context window of 1 million tokens, according to NVIDIA’s developer announcement. Total and active parameter counts describe different aspects of the model and should not be treated as equivalent measures of its size.

Meta released the weights under the Muse model license, with limited commercial-use terms allowing many developers to use the model free of charge. Weight downloads are available through Amazon SageMaker AI, Google Cloud Model Garden, Hugging Face, and Microsoft Foundry Models; API access is available through Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter, and Together AI.

Meta and Apple collaborated to support local execution on high-end Macs, including Mac Studio systems with M5 Ultra chips. Reporting also describes running the model across multiple Macs and an update to Apple’s MLX software toolkit, while NVIDIA’s developer article addresses agentic coding workflows on GB300 NVL72 systems.

Meta chief AI officer Alexandr Wang said Muse Spark Open is not the company’s strongest model, and Muse Wonder remains closed-weight. Meta also acknowledged that Chinese open-weight models outperform Muse Spark Open in software development and some other tasks; the supplied material does not provide the corresponding benchmark scores or testing methodology.