agent
A model running in a loop with tools. It gets a goal instead of a question, asks to run something (read a file, run a command, search), gets the result back, and decides what to do next. The loop is what makes it an agent; the model itself only ever predicts the next token.
Every step re-sends the conversation so far, so token use grows with the square of the number of steps rather than in a straight line. A twenty-step agent run can cost more than a hundred chat messages.
The capability jump people attribute to newer models is mostly this: not that the model reasons better, but that it calls tools reliably enough for the loop to run twenty steps without derailing.