Think Mode: Extended Reasoning for Hard Problems

What Think Mode Does

Most AI questions have a quick path to a correct answer, and a fast response is better than a slow one. But some problems benefit from the model working through the reasoning systematically before committing to an answer. Complex mathematics, multi-step logic, nuanced analysis, and ambiguous debugging problems are cases where rushing increases the chance of error.

Think mode tells Grok to slow down. Before writing its final response, Grok works through the problem internally — evaluating approaches, checking its logic, considering edge cases — and only then produces its answer. The result is slower but measurably more reliable on problems that have a definite correct answer.

How to Turn It On

Think mode is a SuperGrok feature, included in the $30/month plan. Once you have a SuperGrok subscription, a Think toggle appears near the chat input on grok.com. Enable it before sending your message. On mobile, the same control is available in the Grok app.

You can activate Think mode mid-conversation by toggling it before your next message. Each message uses whichever mode is active at the time you send it — there is no need to start a new chat to switch modes.

What Gets Better With Think Mode

Think mode improves performance on tasks where accuracy matters more than speed:

• Complex arithmetic and algebra — Multi-step calculations where small intermediate errors compound into wrong final answers
• Logic puzzles and constraint problems — Any problem where multiple conditions must all be true simultaneously
• Code debugging — Identifying subtle bugs that require tracing logic across multiple functions or understanding side effects
• Structured arguments — Building a case where the conclusion depends on a chain of valid premises
• Medical, legal, and scientific analysis — Domains where a wrong intermediate step leads to a wrong final conclusion

For creative writing, casual questions, brainstorming, or tasks where good-enough is genuinely fine, standard mode is faster and equally useful. Think mode is not better for everything — it is better for the specific category of problems where step-by-step reasoning prevents mistakes.

Reading the Thinking

In Think mode, Grok may display a collapsible "thinking" section before its final answer. This shows the working memory of the model: the steps it considered, the approaches it evaluated, the intermediate conclusions it reached. You can expand this section to understand exactly why Grok arrived at the answer it did.

This reasoning trace is also useful for catching errors. If you spot a wrong assumption or a flawed step in the thinking, you can point it out in your next message — "In step 3 of your reasoning, I think you assumed X incorrectly" — and Grok will reconsider from that point.

Combining Think Mode With DeepSearch

Think mode and DeepSearch address different problems. DeepSearch is for information retrieval: finding out what is currently true. Think mode is for reasoning: figuring out what follows from what is true. For complex analytical tasks that require both extensive research and careful reasoning — say, evaluating a business strategy based on current market conditions — you can activate both simultaneously on SuperGrok.

DeepSearch runs first to gather the current facts. Think mode then applies careful reasoning to those facts. The combination is particularly powerful for decisions where you need both fresh information and rigorous analysis.

Try This

Enable Think mode and ask a maths problem: If I invest $10,000 at 7% annual return compounded monthly, how much will I have after 15 years? Expand the thinking section to see how Grok works through the compound interest formula. Then turn off Think mode and ask the same question — compare both the working shown and the final answer. Next, try a logic puzzle: give Grok a set of constraints (Alice has X, Bob does not have Y, etc.) and ask it to deduce the solution. Notice how Think mode works through each constraint in sequence.