Meta Releases Muse Spark Open-Weight AI Model for Coding and Autonomous Tasks
Meta has released Muse Spark, an open-weight AI model aimed at developers and businesses.
The model supports coding, reasoning, and agentic workflows that can execute multi-step tasks.
The release signals Meta’s continued investment in open AI models as it competes with OpenAI, Google, and Anthropic.
October 16, 2026 - Meta says Muse Spark Open contains 816 billion total parameters, with 42 billion active during inference, and supports a context window of up to 1 million tokens. The model is natively multimodal and targets long-running coding projects, complex software engineering, and multi-step reasoning.
Mac Support and Availability
A notable feature is its ability to run locally on two Apple Mac Studio computers equipped with M5 Ultra chips, using Apple’s MLX framework. This setup requires substantial memory and computing resources, placing local deployment beyond the reach of many standard desktop systems.
Developers can access the model through API platforms including Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter, and Together AI. Meta has not announced a single standard price for access across these providers.
Open Weights, Not Fully Open Source
The release makes the model’s trained weights available for download and adaptation. However, open-weight does not necessarily mean fully open source. Access to model parameters does not automatically include the complete training data, code, or development process.
Meta says Muse Spark Open competes with leading open-weight models, including systems from Chinese developers such as Alibaba and Moonshot AI. Those performance claims are based on company- reported evaluations and should be considered alongside independent testing.
Sources:
https://tech.yahoo.com/ai/meta-ai/articles/meta-launches-open-weight-ai-200453394.html?guccounter=1
