Meta Releases Muse Spark Open AI Model for Coding

Meta Platforms Inc. released Muse Spark Open, an open-weight mixture-of-experts AI model designed for coding, multi-step reasoning, and long-running agentic tasks. The model features 816 billion total parameters, 42 billion active parameters per token, and a one-million-token context window. It is natively multimodal, capable of processing text, images, video, and audio to produce text outputs.

In a partnership with Apple, Meta optimized the model to run on-device using two M5 Ultra Mac Studios with up to 512GB of unified memory each. The weights are available for download and modification via platforms including Hugging Face, Amazon SageMaker AI, Google Cloud Model Garden, and Microsoft Foundry Models. Meta positions the release as a way to compete with leading Chinese models from Alibaba, Moonshot AI, and Z.ai.

This launch follows the release of Meta's closed frontier models, Muse Wonder and Wonder Full. By keeping its most advanced technology closed while providing a less advanced open-weight alternative, Meta aims to convert its AI investments into revenue while supporting the developer community. The company seeks to maintain competitiveness against rivals such as OpenAI and Anthropic.