Skip to main content
Traces record time and cost at both the run level and the span level.

​
In the list

The Duration and Cost columns in the traces list show each run’s wall-clock time and dollar cost. When cost or latency spikes, start here.

​
In the detail

The run summary in the trace detail panel shows Latency, Tokens (the total for the run), and Cost.
  • The span tree shows a duration next to every span, so you can see at a glance which Agent or tool call ate the time.
  • The Metadata section of an Agent span records the model plus tokensIn and tokensOut, so you can pin token usage down to a single call.
If the cost surprises you, open the longest-running Agent span in the tree first. What made the prompt long (system prompt, tool definitions, handed-over context) is right there.