The docs state the mechanism directly: “All questions are evaluated in parallel, so adding more questions to a call typically doesn’t add any latency to the response.”
That flips the usual cost model. With a chat model you ask for the minimum, because every extra field is more tokens to generate and more time to wait. Here the marginal question is close to free in latency terms, so you ask everything any downstream branch might plausibly need, then throw away what the branch you took did not use.
Concrete shape. One call on an inbound support ticket can ask a Choice for which team should handle it, a Score for how frustrated the customer appears, a Noul for whether the message conveys urgency, and whatever a later branch might want. If the team comes back as billing, your code reads the billing answers and ignores the rest. Nothing needs a second request.
Pricing is the other half. Jev charges $0.042 per million input tokens and nothing for output, so speculative questions are billed as input and the answers themselves add nothing. Rate limits sit at 250,000 tokens per second and 1,200 requests per minute.
See the parallel questions fan-out recipe and support inbox triage.