Answers arriving late

Listening and answering are both requests to our servers, so a call needs a network connection throughout. An answer that takes too long is usually the connection rather than the model.

The latency readout in the app is one number — how long the wait ran, from the ask to the first word of the answer. A query limit never switches models or slows a successful answer. Once the allowance reaches zero, new answers are blocked until the reset date or an upgrade.