Models, reasoning, and Deep Think

Home Tools Platforms Gemini Models, reasoning, and Deep Think

Distinguish an available model from an experimental reasoning level for a difficult problem.

Conversation & research · Beta / preview · Documentation checked 2026-10-04

Gemini Apps model and Thinking Level controls.

Deep Think requires adult access and Google AI Ultra or Google AI Ultra for Business. It is experimental and may be suspended or discontinued.

The worked example is illustrative and has not been tested in a signed-in product account.

What it does and how it connects

Distinguish an available model from an experimental reasoning level for a difficult problem.

A model selection and a reasoning level are different controls. Deep Think is an experimental high-reasoning capability, distinct from Deep Research’s investigation process. It can help examine a complex argument, but does not supply external evidence merely because it spends more effort.

Walkthrough

Use the product named in the access panel. The steps below combine documented controls with the editorial example.

  1. Confirm adult access and the documented Google AI Ultra or Ultra for Business entitlement.
  2. On the documented web surface, choose Pro in the model control and then Thinking Level → Deep Think.
  3. Supply a bounded problem with explicit assumptions and constraints.
  4. Inspect the resulting reasoning summary and independently test the important conclusion.

Worked example

Ask for a fair workshop schedule subject to limited rooms and accessibility needs. Provide invented participants and constraints. Validate the proposed schedule against every constraint instead of judging the explanation’s sophistication.

Plan a fictional workshop with two rooms and four sessions. No instructor can teach two sessions at once. The beginner session must precede the agent lab. Show the schedule, list every constraint, and identify anything unsatisfied. Do not assume extra rooms or staff. Give a simple method to check the plan.

Inspect and verify the result

Enumerate constraints and check the schedule directly. Ask for a counterexample or alternative plan. If a constraint was omitted, revise the input rather than accepting a fluent justification.

Limits and troubleshooting

Deep Think is experimental, may be suspended, and has account usage limits. It is not the default available to every user. If missing, verify entitlement and rollout; an ordinary available model can still handle a smaller clearly specified problem.

Data, permissions, and actions

The problem and supplied context enter Gemini Apps under the account’s relevant terms. No external service is needed for this synthetic exercise. Adding tools or private files would create a separate permission and data decision.

Deep Research and source selection

Review an investigation plan and deliberately choose web, uploaded, or connected sources.

Explore Deep Research and source selection

From research to a Canvas handout

Turn a verified public-source report into a beginner handout and an inspectable prototype.

Explore From research to a Canvas handout

A notebook and weekly learning review

Keep approved lesson context organized and generate a private recurring review.

Explore A notebook and weekly learning review

Evaluations: does the system do the job?

Create representative tests for quality, tool behavior, refusal, escalation, and the effects of a change.

Explore Evaluations: does the system do the job?

Sources and verification

Documentation checked 2026-10-04. Examples are illustrative unless described otherwise.

Deep Think: experimental reasoning · Google

Share this page