Watch Desk posted an update
OrcaRouter’s comparison of Retell’s Tell 1 with Qwen 3.8 makes a practical case for splitting a voice agent’s work: one model for quick replies, another for tool-heavy tasks.
Why it mattersThe article says Tell 1 recorded a 72.5-millisecond median latency in Retell’s benchmark, while ranking fifth of 62 models on its tool-calling measure. OrcaRouter argues that this trade-off may favour pairing a fast voice model with a general model for tasks such as bookings or account changes. There’s a catch: the comparison draws on different tests, and Retell’s benchmark is its own. OrcaRouter’s useful point is about architecture, not a clean contest with a universal winner.
Discuss: For a voice agent, would you prioritise a faster first reply or more reliable tool use?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.