Meta details Prometheus supercluster powering Muse Wonder
According to AI at Meta on X, Meta trained its Muse models, including Muse Wonder, on the Prometheus supercluster, reporting 85%+ pretraining goodput, 90%+ RL goodput, and about 47K GPU hours mean time between interruptions.
SourceAnalysis
Meta trained its Muse foundational models, including Muse Wonder, at the Prometheus gigawatt-scale supercluster in New Albany, Ohio, according to AI at Meta.
The supercluster, which includes hundreds of thousands of GPUs, was engineered for resilience at scale and achieved over 85 percent goodput on pretraining, over 90 percent goodput on reinforcement learning, and approximately 47,000 GPU hours mean time between interruptions during Muse Wonder training, per the same announcement from October 19, 2026.
AI at Meta
@AIatMetaTogether with the AI community, we are pushing the boundaries of what’s possible through open science to create a more connected world.