Deep research

Background job · minutes, not seconds

A longer, structured version of research mode for questions where you want a report rather than an answer.

How a deep research run goesA planner model splits your question into facets. One agent per facet fetches pages and extracts claims with quotes. If agents report gaps and the run is under four rounds and 90 minutes, the planner adds facets and gathering repeats. Then the report is written from numbered sources, and every citation is checked. A failing report is regenerated once. The cited report lands in your chat and under Jobs.yes, plan more facetsnoyesnoYour question(Pro, Max or Ultra)1. Plansub-questions, facets,success criteria2. Gather, one agent per facetfetch up to 6 pages each,extract claims with quotes3. Gaps left?under 4 rounds,under 90 minutes4. Write the reportfrom numbered sourcesEvery [n] resolves,every finding cited?Regenerate onceCited report in your chatand under Jobs(warning line if issues remain)
Gathering can repeat for up to four rounds; the citation check runs once at the end, with one repair.
Diagram as text
  1. You start the run from a chat on Pro, Max or Ultra. It is a background job.
  2. Plan: a strong model writes sub-questions, a set of facets and the success criteria.
  3. Gather: one agent per facet fetches up to six pages and extracts claims, each with a short quote.
  4. If agents report open gaps and the run is under four rounds and the time limit (90 minutes by default), the planner proposes one to three more facets and gathering repeats.
  5. Otherwise the report is written from the numbered sources.
  6. The citation check: every [n] must point at a fetched source and every key finding must carry a citation.
  7. If the check fails, the report is regenerated once. If problems remain, the report ends with a warning line.
  8. The report is written into your chat and appears under Jobs.

How a run goes

  1. [1]

    Plan

    A strong model turns your question into the pieces the run works from.

    • Sub-questions.
    • A set of facets to investigate.
    • The criteria that would mean the question has actually been answered.
  2. [2]

    Gather, in parallel

    One agent per facet

    Facet

    Facet

    Facet

    and so on

    Each agent finds sources, fetches them, and pulls out specific claims with the quote each one rests on.

  3. [3]

    Decide whether to go again

    If the agents report open gaps, the planner proposes one to three more facets and the run gathers again. It stops when one of these is true:

    • No gaps are left open
    • It has run for 90 minutes (the default time limit)
    • It has done four rounds
  4. [4]

    Write the report and check the citations resolve

    The report is written from the numbered sources. Every [n] in it must point at a source that was actually fetched, and every key finding must carry at least one.

    • [1] a source the run fetched holds
    • [2] a source the run fetched holds
    • [3] nothing that was fetched fails
    • [ ] a key finding carrying no citation at all fails

    Either failure sends the report back. It is regenerated once; if a problem remains, the report ends with a warning line saying how many citation issues could not be resolved.

The dashed return is the round that goes again: it starts back at gathering, with the extra facets.

What comes back

The report

Citations you can follow.

Alongside it

A record of what was dropped along the way.

It appears under Jobs like other long-running work, and survives the browser being closed.

What it costs

Several models ×many fetches ×multiple rounds

This is one of the more expensive things you can run. You cannot set a spend cap for a run: it is billed from your credit like any model use and stops at zero credit, after four rounds or after the time limit. Check your balance before an open-ended question. Deep research needs Pro, Max or Ultra.

Not the same guarantee as research mode

Research mode

Checks
That a quote appears verbatim in the page it cites.
Strength
The stronger check.
Reach for it when
A single claim has to be bulletproof.

Deep research this page

Checks
That every citation resolves to a source it really fetched, and that nothing is left uncited.
Strength
A weaker guarantee, on far more material.
Reach for it when
You need breadth and will read the sources yourself.