Extended Thinking

Sometimes your assistant needs to think through a problem before responding. Extended thinking makes that process visible — you can watch the assistant's reasoning unfold in real time, then review it after.

How it appears

When extended thinking activates, you'll see a collapsible section above the assistant's response:

• While thinking — A brain icon with a shimmer animation and the label "Thinking..." appears. The section auto-expands so you can follow the reasoning as it streams in.
• After thinking — The label changes to "Thought for X seconds" and the section auto-collapses after about a second, leaving the final response visible.

The thinking content supports rich formatting — code blocks, math equations, diagrams, and standard markdown are all rendered inside the thinking section.

Interacting with thinking

You can expand or collapse the thinking section at any time by clicking it. If you manually expand it while the assistant is still thinking, it stays open — the auto-collapse won't override your choice.

This is useful when you want to understand why the assistant gave a particular answer, or when you want to verify its reasoning on a complex question.

When it activates

Extended thinking doesn't appear on every response. It's most likely to show up when the assistant is working through:

• Complex questions — Multi-part questions that require weighing several factors
• Analysis and reasoning — Tasks like comparing options, evaluating trade-offs, or working through logic
• Nuanced topics — Questions where the answer depends on context and the assistant needs to consider multiple angles

Simple factual questions or short tasks typically get a direct response without visible thinking.