Comment by paragagrawal

In agentic use cases, we save on end-to-end latency by spending more time and compute on individual searches. This happens because agents do fewer searches, use fewer tokens, and end up using fewer thinking tokens when using the Parallel Search API.