Meta Launches Muse Spark Open: 816B Coding Model with 1M Token Context
What happened?
Meta announced Muse Spark Open, an agentic coding model built on a mixture-of-experts (MoE) architecture with 816B parameters, 42B active, and a 1M token context window. It was trained on NVIDIA GB300 NVL72 with Blackwell Ultra, according to the NVIDIA Developer blog.
FAQ
What is Meta's Muse Spark Open?
It is an agentic coding model built on a mixture-of-experts (MoE) architecture with 816B parameters, 42B active, and a 1M token context window, optimized for agentic coding tasks and codebase understanding.
How does it compare to other coding models?
It combines a large MoE footprint with fewer active parameters, a 1M token context, and training on NVIDIA GB300 NVL72 Blackwell Ultra, boosting first-attempt accuracy and tool-calling reliability versus smaller-context models.
Should MENA teams adopt it now?
It depends on infrastructure readiness; the model targets advanced GPU platforms like GB300 NVL72, so regional enterprises may need cloud partnerships or compatible data centers before full adoption.
Why does a 1M token context matter?
It allows loading entire codebases or large repositories into a single context, reducing task splitting and improving cross-file understanding.
Source: NVIDIA Developer (AI)
AI-assisted summary of the source above. High-confidence stories publish automatically; others wait for review. Facts that are not in the source are not added.