Model
Chooses which underlying AI model answers your questions. The
recommended default is a good all-round choice. Switching models can change answer quality,
speed, or cost, but usually isn't necessary — change it only if you're comparing models or
troubleshooting a specific problem.
Thinking level
Controls how much the model reasons privately before
answering. Higher levels can improve accuracy on hard questions — like comparing several
videos or working out a timeline — but make answers slower to start. Lower levels are useful
when you want a quick answer to a simple question and don't need deep analysis.
Agentic video processing
When on, the model actively re-watches specific parts of
a video — rewinding to a segment, zooming into frames, or re-listening to audio — while
forming its answer. This is useful for detailed or hard-to-see footage. Turn it off for a
faster, static analysis at a fixed frame rate; that's usually enough for simple questions
and short videos, but it can miss brief or subtle details.
Cite timestamps in answers
When on, answers include timestamps pointing to the
moment in the video being described, and each one is a button that jumps the player to that
moment. Useful when you want to check a specific moment for yourself. Turn it off if you only
want a plain-language answer without timestamp references — for example, when listening with a
screen reader and the extra numbers add noise. Note: with this off there are
no timestamps to click, so jumping to a moment in the video is unavailable.
Changes apply to your next message.