Meta completes first phase of Prometheus supercluster, trains Muse on it
What happened
As reported on October 19, 2026, the first phase of Meta's gigawatt-scale Prometheus supercluster is complete and has been used to train the Muse model family. The cluster currently has hundreds of thousands of GPUs and is planned to scale to about 1 million. Muse Wonder pretraining ran at over 85% effective throughput, and RL training at over 90%. After a year of engineering work, cluster interruptions fell 99% from their peak. Scaling to about 1 million GPUs remains the plan.
Written by AI from the coverage · updated 2 hours ago
Coverage
Follow the reports to see the story from different sides.
- AI HOT · Tips & opinionsPickMeta details gigawatt-scale Prometheus supercluster; first phase is training Muse models
Meta says the first phase of its gigawatt-scale Prometheus supercluster is complete and is being used to train the Muse model family. It has hundreds of thousands of GPUs now, with plans to reach about 1 million. Effective throughput tops 85% for Muse Wonder pretraining and 90% for RL training, and a year of engineering work cut cluster interruptions 99% from their peak.
Heat over time
Not enough continuous observations to draw a trend yet.