Meta Launches 816B-Parameter Open AI Model

Meta has launched Muse Spark Open, an 816-billion-parameter open-weight multimodal model designed for coding and agentic tasks.

What happened

The model has a one-million-token context window and is positioned by Meta as its most capable US open-weight system. It can run locally on sufficiently powerful Apple hardware, although deployment requires roughly one terabyte of memory, putting it primarily in high-end workstation territory. Meta says it has tested the model with Muse Code on software-development and video-editing workflows. Its benchmark claims are company-reported and should not be treated as independently established superiority over competing systems. The launch is a model release rather than a funding event.

Why it matters

Most frontier-capability models remain tied to cloud infrastructure controlled by a small number of providers. Open weights give developers more control over deployment, data handling and customisation, while local execution can reduce dependence on remote APIs for sensitive workloads.

The bigger picture

The release intensifies competition in open-weight AI and tests whether extremely large models can find practical local use despite demanding hardware requirements. The market may increasingly segment between compact models for ordinary devices and massive open systems for specialised workstations and private infrastructure.