The model runs on Yugam's own server in Bangalore. Roughly 2 words/sec — answers stream in as they're written.