A note before the map. OpenAI renames and reshuffles these regularly, and the allowances on each plan shift, so treat the labels as a snapshot and the logic as the durable bit: a fast mode for everyday work, a slow mode for hard work, and a handful of tools for specific jobs. That logic survives every rename.
Somewhere near the message box, ChatGPT has quietly grown a set of switches. Instant, Thinking, Pro, something called Deep Research in a tools menu, and a Canvas that appears sometimes. Most people, entirely reasonably, ignore all of it and type into whatever was already selected, wondering occasionally whether they are using the thing wrong.
Mostly you are not. But two or three of these switches are genuinely worth knowing, because they are the difference between a shallow answer and a considered one, and between ten browser tabs of research and one report.
ChatGPT has three speeds and a toolbox. Auto picks the speed for you and is the right default. Instant is the fast everyday mode, and Thinking works slowly through hard, multi-step problems. Pro is the heavyweight version on the top plans, for work where being right matters more than being quick.
The toolbox holds Deep Research (a properly cited report, in minutes), Canvas (a side-by-side editor for drafts) and Voice. Use the speeds for how hard the question is, and the tools for what kind of job it is.
The three speeds
| Mode | What it is | Use it when |
|---|---|---|
| Instant | The fast default, answers in seconds. | Everyday work: drafts, summaries, rewrites, quick explanations. |
| Thinking | Works step by step, takes noticeably longer. | Multi-step problems: analysis, planning, tricky comparisons, anything with logic in it. |
| Pro | The deepest and slowest, on the top-tier plans. | High-stakes work where accuracy beats speed. Most people never need it. |
The mental model is a colleague. Instant is them answering off the top of their head, which is ideal for most things. Thinking is them saying "give me a minute" and actually working it through, and you can often watch the steps. Pro is them clearing the afternoon for it.
The mistake is running everything on Instant out of habit and then blaming the tool when a genuinely hard question gets a shallow answer. The question was fine; the speed was wrong.
Auto sits above all three and picks for you based on the question, which is why it is the right default. Flick the switch manually only when you know the job is hard.
The toolbox
Deep Research is the one worth learning this month. Instead of answering from memory, it goes off for several minutes, reads a large number of sources across the web, and comes back with a structured, cited report. For "what should we know about this market", "compare these options properly", or anything you would otherwise spend an afternoon of browser tabs on, it is the single biggest time-saver in the product.
It lives in the tools menu and comes with a monthly allowance on paid plans, but the sources still want opening before you quote them. Cited is not the same as correct, and the checking method is the same as for any other output: how to fact-check AI output.
Canvas turns the chat into a side-by-side editor, your document on one side and the conversation on the other, so you can rework a paragraph without regenerating the whole piece. Use it the moment something goes past a few rounds of edits, and it stops the maddening loop of asking for one change and getting back a different document.
Voice is for talking a problem through hands-free, on a walk or a commute. It shines for thinking out loud and brainstorming, less so for anything you need in writing.
Search, for completeness, is the tool that checks the live web for current information, and ChatGPT often uses it on its own when a question needs today's facts rather than memory.
The trap nobody mentions
On the free plan, the fast lane has a cap, and when you hit it mid-conversation, ChatGPT quietly drops you onto a smaller, weaker model without much ceremony. So if the tool seems brilliant in the morning and dim by the afternoon, that is often what happened, not your imagination.
It is one of the reasons ChatGPT can appear to change quality from one hour to the next, and it is worth knowing before you conclude the whole thing has got worse. There is more on that drift in why ChatGPT sounds different this week.
The habit that covers most of it
You do not need to memorise the map. Leave it on Auto, and build one reflex: when a question is genuinely hard or genuinely matters, take it out of the fast lane, switch to Thinking, or hand it to Deep Research if the job is research.
That single habit, matching the effort of the tool to the weight of the question, is most of what separates the people who get real value from these switches from the people who never noticed them. And whichever mode produced the answer, the checking rules do not change. A slow, well-reasoned wrong answer is still wrong.
The modes at a glance
| Switch | What it is for | When to reach for it |
|---|---|---|
| Auto | Picks the speed for you. | Leave it here as the default. |
| Instant | Fast everyday answers. | Drafts, summaries, quick rewrites. |
| Thinking | Slower, step-by-step reasoning. | Hard, multi-step or high-stakes questions. |
| Pro | Deepest reasoning, top-tier plans. | Rare, when accuracy really has to beat speed. |
| Deep Research | Cited, structured report from the live web. | Anything that would otherwise be an afternoon of tabs. |
| Canvas | Side-by-side document editor. | Drafts you are editing over several rounds. |
| Voice | Hands-free conversation. | Thinking out loud on a walk or a commute. |
Not sure where to start?
The right mode helps. The right tool helps more.
The free finder tells you which AI to open for the task in front of you, in about 60 seconds, and the full "AI, sorted." reference lands in your inbox.
Try the free finder →One question in, one answer out.
Clair helps non-technical professionals know when to trust their AI, when to check it, and when to skip it.