01 / The Model
Model provider request
One round trip: everything the model should see goes out, one reply comes back. The unit that cost and waiting time are actually made of. A single instruction from you can trigger dozens.
How you’ll hear it
“It was one question. Why so many calls?”
“An agent makes a request for every step it takes.”
Avoid
Assuming one question equals one request.
The longer version
A request is a single round trip: everything the model should see goes out, one reply comes back. Agents make many of these per task, one per step, which is why their cost and delay grow in ways a simple chatbot’s do not.