The prompt

A question, a depth, a citation rule

Research the current annual price of the three
password managers I have open, then tell me which
is cheapest per year and what each includes.
Cite every price.

Pick Quick for a first look, Standard for most decisions, Deep when the answer rests on several sources that disagree.

What happens

Plan. Two to four search queries, written to cover the question from different angles.
Search. DuckDuckGo by default, with another engine tried if the first shows a CAPTCHA or nothing useful.
Read. The best results, different sites first, open in parallel background tabs of your browser. JavaScript, cookies and your logins all apply.
Fill gaps. Standard and Deep runs look for what is missing, or what rests on a single source, and search again.
Write. The report answers first, then gives detail. Every claim carries a numbered source, and citations to sources that do not exist are removed.

Tabs are closed when it finishes. The report stays in the chat and in History, and can be copied, downloaded as Markdown, or repeated on a schedule.

What to check

Open the numbers. Click a citation before you rely on it. The link should be the page the claim came from.
Look at the date. A price from last year is still a source; it is just not the current price.
Check the depth. A Quick run is a first look, not a decision document.

What can go wrong

The best sources are PDFs. Research cannot read PDFs yet. It will lean on HTML sources or return a thinner answer.
Every source is behind a login you do not have. It will report the gap instead of inventing an answer.
A site blocks automation. The run skips it and notes why; another source takes its place when one exists.
The question has no answer. It says so. That is the intended behaviour, not a failure.