A "turn" is a single model generation (one call to the model that produces an assistant message). Turn limits are distinct from message limits, which count all messages in the conversation (user, assistant, and tool messages). One turn often results in several messages (e.g. an assistant message plus the tool messages from its tool calls).

A turn is recorded once per top-level generate() call (after retries and fallbacks have resolved, and including cache hits). Generations made via model.compact() do not count as turns. Turn limits are checked whenever a turn is recorded.