# How do you use Grok 3's Think and DeepSearch modes?

Date: 2025-02-26 · Author: Robert Scales · https://guidetoartificialintelligence.com/en/learn/use-grok-3s-think-deepsearch-modes/

**Answer:** Press Think when a question needs careful step-by-step work, such as math, logic, code or a plan with constraints, and press DeepSearch when the answer depends on current information from the web and X that Grok has to gather and summarize. Use plain Grok 3 for everything else, and check the sources DeepSearch shows before you rely on its report.

Grok 3, the model family xAI unveiled on February 17, 2025, gives you three ways to ask a question in the Grok app, on grok.com and on X. You can send it normally, you can press **Think**, or you can press **DeepSearch**. They are not three levels of quality. Each one fits a different kind of question, and picking the right one saves time and gives better answers.

Access depends on your plan. xAI says X Premium+ subscribers have Think and DeepSearch right away, Premium users get Grok 3, and other Grok users get the features with usage limits. TechCrunch reports that a new SuperGrok subscription adds more reasoning and DeepSearch queries.

## What each mode does

**Plain Grok 3** answers directly from what the model learned in training. It is the fastest choice for drafting, explaining, summarizing text you paste in, or brainstorming.

**Think** turns on Grok 3's reasoning mode. A **reasoning model** works through a problem in steps before it answers, trying approaches, checking its work and backing up when it finds a mistake. xAI says Grok 3 (Think) can spend from a few seconds to several minutes this way and that you can open its reasoning to inspect it. TechCrunch notes that some of those "thoughts" are hidden in the app, and that a more demanding setting called Big Brain uses extra computing for the hardest questions. xAI recommends reasoning for math, science and programming.

**DeepSearch** is what xAI calls its first agent: a program that takes several steps on its own. It searches the web and X, reads what it finds, tries to reconcile conflicting claims and ends with a summary report.

| Your question | Best mode |
|---|---|
| "Rewrite this paragraph in a friendlier tone" | Plain |
| "Which of these three loan offers costs least over five years?" | Think |
| "Why does my spreadsheet formula return an error?" | Think |
| "What changed in my city's recycling rules this month?" | DeepSearch |
| "What are people on X saying about this product launch?" | DeepSearch |

## A worked research task

Suppose you run a small bakery and are considering a switch to compostable packaging. Here is how the two special modes can split the work. The prompts are written so you can reuse them; any output you get will vary.

Step 1, gather current facts with DeepSearch:

```prompt
Find current information on compostable food packaging for small bakeries in Canada: typical suppliers, approximate price differences compared with plastic clamshells, and any municipal rules on what composting facilities accept. List each source with its date. Say clearly where sources disagree or where you found no reliable information.
```

Step 2, open the sources it lists. Check that each one exists, is recent, and says what the summary claims. Remove anything you cannot confirm.

Step 3, reason over the confirmed facts with Think:

```prompt
Using only the figures below, compare the yearly cost of plastic and compostable packaging for a bakery selling 120 boxed items a day, six days a week. Show each step of the calculation, then list which assumptions most affect the result.
[paste the figures you confirmed]
```

Splitting the task this way plays to each mode's strength: DeepSearch finds and summarizes, Think calculates and weighs, and you check what sits in between.

> **Try it:** Ask the same question three ways: plain, Think and DeepSearch. Pick something with a factual part and a reasoning part, such as "Is it cheaper to take the train or drive from Ottawa to Montreal this weekend?" Note what each mode got right, what it missed, and how long it took.

## Habits that help

Give Think a complete problem. Reasoning models do best when all the numbers and constraints are in the prompt; they cannot reason their way to facts you left out. Give DeepSearch a bounded question with a place and a time frame, and ask it to name its sources and flag disagreements. And keep in mind that xAI describes the models as still in training and in beta, so behavior may change from week to week.

## How does it compare with ChatGPT for research?

OpenAI's equivalent is **deep research**, launched in ChatGPT on February 2, 2025. OpenAI describes it as an agent that browses for 5 to 30 minutes, can read text, images and PDFs, and produces a longer report with citations; you get a notification when it finishes. At launch it was offered first to Pro subscribers, with up to 100 queries a month. Grok's DeepSearch is pitched as a faster agent that ends with a concise report, and it also searches posts on X, which helps when the question is about public reaction or very recent news.

So the choice depends on the job. For a quick, current overview or a pulse of what people are saying, DeepSearch is the lighter tool. For a long, structured report on a complex subject where you want many cited sources, ChatGPT deep research is designed for that depth.

Both share the same weakness. OpenAI states that deep research can make up facts, draw wrong conclusions and struggle to tell authoritative sources from rumor, and an X-heavy search can amplify rumor too. Whichever you use, treat the report as a well-organized first draft: open the key sources, confirm the numbers, and keep only what you have checked.

## Sources

- Grok 3 Beta — The Age of Reasoning Agents: https://x.ai/news/grok-3
- Elon Musk's xAI releases its latest flagship model, Grok 3: https://techcrunch.com/2025/02/17/elon-musks-ai-company-xai-releases-its-latest-flagship-ai-grok-3/
- Introducing deep research: https://openai.com/index/introducing-deep-research/