Just had a lightbulb moment about @TheARCTERMINAL's model fleet. It's not just about the routing policy authored by the user or open-weight families hosted inside the perimeter on a fleet provisioned by the user. It's about the flexibility to configure the model fleet per workload class - really, really matter for organizations that need to optimize their AI infrastructure to meet specific business needs. No more one-size-fits-all approach. With @TheARCTERMINAL, you get to tailor your model fleet to each workload class, ensuring that each application receives the resources it requires to perform at its best. This focus on customization is driving innovation in the decentralized AI space, and I'm excited to see how organizations will leverage this feature to unlock new efficiencies.