Perplexity's hybrid orchestrator auto-routes each task between device and cloud.

At Computex, Perplexity introduced what it calls the first hybrid local-server inference orchestrator, routing each task to on-device or cloud frontier models without the user choosing in advance, shipping to Perplexity Computer in July. Automatic local/cloud routing is the architecture to watch as on-device models close the gap. It's the same heterogeneous-routing idea from the MiniMax story, just pushed down to the OS level and made invisible.