--- id: chatgpt-models-thinking-and-web-search lang: en track: chatgpt title: "Models, thinking, and web search" seoTitle: "ChatGPT thinking levels, models & web search" description: "How ChatGPT's thinking levels work, when to use Instant or a higher level, and how web search and deep research bring in current facts you can check yourself." part: getting-started partTitle: "Getting started with ChatGPT" partOrder: 2 order: 3 level: Beginner durationMinutes: 4 outcome: "Choose the right thinking level for a task, and use web search and deep research well." summary: "Match the setting to the task: Instant for everyday work, a higher thinking level for hard problems, web search for anything current, and deep research for a cited report." draft: false updated: "2026-10-14" furtherReading: - title: "GPT-6 and other models in ChatGPT" url: "https://help.openai.com/en/articles/20001354-gpt-6-and-other-models-in-chatgpt" note: "The current models and which thinking levels each plan gets." - title: "Searching the web with ChatGPT" url: "https://help.openai.com/en/articles/9237897-searching-the-web-with-chatgpt" note: "Turning search on, and reading citations and sources." - title: "Deep research in ChatGPT" url: "https://help.openai.com/en/articles/10500283-deep-research-in-chatgpt" note: "How a research plan becomes a cited report." - title: "ChatGPT release notes" url: "https://help.openai.com/en/articles/6825453-chatgpt-release-notes" note: "Where model and picker changes are announced." keywords: - ChatGPT models - ChatGPT thinking - ChatGPT Instant - ChatGPT web search - ChatGPT deep research - which ChatGPT model to use quiz: intro: "Five quick questions. Ace them all and the next lesson unlocks." passScore: 5 questions: - q: "For most everyday work, drafting, summarising, quick questions, which setting should you use?" options: - "The highest thinking level, for the best quality" - "Instant, the fast default" - "Deep research, so every answer is checked" - "Web search, so answers are always current" answer: 1 explain: "Instant handles the bulk of daily work quickly. Higher thinking levels are slower and use more of your allowance, so save them for problems that need them." - q: "You have a tricky, multi-step problem to reason through, and it needs no current information. What helps most?" options: - "Turn on web search for more information" - "Ask the same question several times and compare the answers" - "Start a Temporary Chat so nothing distracts it" - "Pick a higher thinking level so it reasons first" answer: 3 explain: "A higher thinking level makes ChatGPT work through the problem step by step before replying. That pays off on hard reasoning, planning, and analysis." - q: "You need this week's prices and news for a short market update. What do you use?" options: - "Web search, then check the cited sources" - "A higher thinking level, so it thinks harder" - "Memory, so it recalls last week's update" - "Pro, which is trained on more recent facts" answer: 0 explain: "Built-in knowledge stops at a cut-off date. Web search fetches current pages and shows citations you can open, so you can check the facts yourself." - q: "When is deep research worth its wait?" options: - "For a quick factual question you need answered right now" - "For polishing an email before you send it" - "For a multi-source question that needs a cited report" - "For any question where accuracy really matters" answer: 2 explain: "Deep research plans the work, reads many sources, and writes a report with citations. It takes minutes, not seconds, and counts against a monthly allowance, so use it for questions that deserve a report." - q: "Your Business workspace answers on Instant, but sometimes thinks longer on its own. Why?" options: - "A higher level is switched on for everyone by default" - "'Higher intelligence' can add thinking automatically" - "Memory tells it to think longer on familiar topics" - "Instant switches to Pro when the chat gets long" answer: 1 explain: "On Business, a setting called Higher intelligence (Settings, General) lets ChatGPT use extra thinking when Instant is selected and a request looks hard. You can switch it off if you prefer predictable speed." pass: badge: "Perfect score" title: "You can pick the right setting now." body: "Full marks. You know when to stay on Instant, when to make ChatGPT think, and when to send it to the web. Next up: Pages and Sites, where answers turn into work you can keep." retry: "Check the marked answers and give it another go. You've got this." --- > **Level:** Beginner ยท **Reading time:** ~4 min > > **What you'll be able to do:** pick the right thinking level for a task, and use web search and deep research well. ChatGPT gives you a few dials. The good news: **the defaults are sensible, so you can ignore all of this and still do good work.** But knowing what the dials do turns a slow, so-so answer into a fast, sharp one. There are three worth learning: how hard ChatGPT thinks, whether it searches the web, and when to hand it a full research job. ## Models and thinking levels Behind ChatGPT sits a **model**, the engine that writes the answers. OpenAI renames and replaces models often (at the time of writing, the current one is called GPT-6), so do not memorise names. Learn the dial instead. That dial is the **thinking level**, set from the **model menu** in the message box. It runs from fast to thorough: **Instant** answers straight away. It is the default for good reason: drafting, summarising, rewriting, and everyday questions. If you are not sure, stay here. **Medium and High** make ChatGPT **reason step by step before it answers.** Slower, but much better on problems with real teeth: a knotty calculation, a plan with many moving parts, weighing a hard decision. Think of a colleague who says "give me a minute" and then gets it right. **Extra High and Pro** are for the hardest jobs, the kind where you would happily wait a few minutes for a careful answer. On Business you have all of these; on Plus, Instant, Medium, and High.
Pick the dial for the job The thinking levels in a row from fast to thorough: Instant for everyday work, Medium and High for harder problems, Extra High and Pro for the hardest jobs. Higher levels take more time and more usage. Below, web search for anything current and deep research for a cited report. Pick the dial for the job Stay on Instant for most things. Raise the level when a problem has teeth. Instant Medium High Extra High Pro everyday work Medium and High: harder problems Extra High, Pro: hardest jobs Faster More thorough, more time, more of your usage Web search For anything current. Answers come with sources: open them. Deep research For a question that needs a report. Plan, many sources, cited report.
Higher levels take longer and **use more of your allowance**, so cranking everything to the top is the most common beginner mistake. It just makes you wait. You can switch level **in the middle of a chat**; it applies to the next reply. One Business detail: a setting called **Higher intelligence** (in Settings, General) lets ChatGPT add extra thinking on its own when you are on Instant and a question looks hard. Leave it on for convenience, or switch it off if you want Instant to stay instant. ## Web search: for anything current A model learns from data up to a **cut-off date**. It does not know this morning's news, this week's prices, or a policy that changed yesterday. For that, ChatGPT uses **web search**. Often it searches by itself when a question needs current facts. You can also ask for it: pick **Search** from the tools behind the **+** button, type **/** and choose Search, or simply say "search the web for...". The answer comes back with **citations**, small links you can open, and a **Sources** list. One rule keeps you safe: **trust, but verify.** OpenAI's own help pages warn that results can be incomplete, outdated, or wrong. On anything that matters, open the source and check the date. Let ChatGPT do the finding; you own the fact. What does checking look like? Say ChatGPT tells you a supplier raised its prices by 8% this month. Open the citation. Is it the supplier's own announcement, or a blog repeating a rumour? Is it from this month, or from last year? Does the page actually say 8%? Thirty seconds, and you know whether the sentence can go into your report. If the source is weak, ask ChatGPT to search again and name the source you trust. ## Deep research: when you need a report For a bigger question, the kind where you would normally spend an afternoon with twenty browser tabs, there is **deep research.** You describe what you want to know, ChatGPT proposes a **research plan** you can edit, and then it works through many sources and writes a **structured report with citations**, which you can download as a Word, PDF, or Markdown file. It takes minutes, not seconds, and each plan gets a set number of deep research tasks a month, so save it for questions that deserve a report: a market scan, a comparison of suppliers, a summary of new regulation. On Business it can also read from connected apps such as SharePoint or Google Drive, read-only, within your own access rights. ## Matching the setting to the task Put it together and a simple habit falls out. **Stay on Instant for most things.** When a problem is genuinely hard and self-contained, **raise the thinking level.** When you need something current, **search**, and check the sources. When you need a proper written answer from many sources, **run deep research.** They combine, too: a thinking level and web search work fine together. ## Try it Take one real question with a clear right answer, a bit of logic or a calculation from your work. Ask it on **Instant**, then ask it again on **High**. Compare: the thinking answer usually shows its working and catches a detail the quick one missed. Now ask something about *this week*, a price or a recent change in your sector, and make sure **Search** is used. Open two of the citations and check them. Two short tests, and the dials stop being a mystery. *(General exercise. Role and industry versions come once accounts are in.)* ## Key terms [Model](/glossary#chatgpt-model), [Thinking level](/glossary#thinking-level), [Web search](/glossary#chatgpt-web-search), [Deep research](/glossary#deep-research), [Token](/glossary#token)