Agents

AI agent reasoning and tool selection tests planning, tool choice, and self-diagnosis capabilities. These skills are crucial for autonomous ticket triage systems and complex agent workflows where models must decide which tools to use, plan multi-step processes, and diagnose when things go wrong. This represents some of the most challenging tasks as they require open-ended reasoning and decision-making.

Input Tokens

Output Tokens

Duration

Cost

Context

Input

USER INTERRUPTION: The user says, "Actually, never mind about the reminder — just make sure the company dinner is on my calendar."  What do you do to adapt your plan?

Expected output

I will revise the plan by removing the reminder step entirely, then proceed to: (1) search and read the relevant email to obtain the dinner date, (2) check for calendar conflicts, (3) create the event with calendar.create_event **without** the reminder_days_before parameter, and (4) inform the user of any conflicts and confirm the event has been added.  What matters is that the plan is explicitly updated and no reminder is scheduled.

Model output

Reasoning: The user changed their request to remove the reminder portion and only wants the event on the calendar. I will adapt the previous plan by removing the step about setting a reminder. I will continue with the steps to find the event details by searching emails and reading the specific message for the date. Then I will create the calendar event without adding a reminder. Finally, I will confirm there are no scheduling conflicts for that date.

Response: Understood. I will skip setting a reminder. I’ll search your emails for details about the company dinner, extract the date, and then add the event to your calendar without any reminders. I will also check for schedule conflicts on that date.