Guides › New to AI

Which AI model, and how hard should it think?

Updated Oct 26, 2026. Settings checked against each company's help pages, linked below.

Most AI apps let you choose a model and a thinking level. Here's when to change them, in 60 seconds.

What's the model picker?

A model is the AI "brain" an app uses. Companies offer several: fast, light models for everyday questions and bigger, smarter models for hard work. The model picker is the menu where you choose. It's usually near the message box.

Why have several? Bigger models are slower and cost the company more to run, so they're limited, and the best ones come with paid plans. Fast models are fine for most everyday things.

When is the default fine? Most of the time. Grok, Gemini, ChatGPT, Copilot and Perplexity all have an automatic option that picks for you (called things like Auto, Smart or Best).

What's "thinking" or "effort"?

Many apps also let the AI think longer before it answers. The setting is called thinking, effort or reasoning, and sometimes it's a mode like "Expert" or "Think Deeper". More thinking means:

Low / Instant
fastest, uses least

Medium / Standard
everyday

High / Extended
hard problems

Max / Pro / Heavy
slowest, uses most

Our diagram, not a screenshot. Level names differ by app.

Better answers on hard, multi-step problems (maths, planning, coding, tricky questions).
Slower answers: seconds to minutes.
Your limit runs out faster. Anthropic and Google both say higher levels use more of your allowance (Claude help⁵, Gemini help⁸).

For a simple question, extra thinking doesn't help much. It just takes longer.

If this, pick that

Cheat sheet: which model and thinking level for each kind of task
If you want to…	Pick	Why
A quick question, a fact, a recipe, a short email	Default or Auto, fast / no extra thinking	It's quick and saves your limit.
Rewrite or tidy up something you wrote	Default, fast	Writing help rarely needs deep thinking.
Summarize or ask about a long document	A smarter model (the plan's main or Pro model)	Bigger models read long files better, and paid plans read longer files.
A tricky problem: maths, a puzzle, planning with many steps	Higher thinking (High, Extended, Expert, Think Deeper)	Thinking longer means fewer mistakes on multi-step problems.
Coding	A smarter model, medium to high thinking	Coding goes better with the stronger models; save the top level for big jobs.
An important decision (health, money, legal)	The smartest model with high thinking, then check with a professional	Better reasoning, but still not a substitute for an expert.
A report with sources	The app's Deep Research mode	It searches many websites for you; it can take several minutes.
You're running low on your limit	A faster model, lower thinking	Faster models and lower levels use less of your allowance.

Where the setting is in each app

From each company's own help pages. Some options need a paid plan.

Where to find the model and thinking settings
App	Where to find it	Your choices	Good to know	Source
Grok	Mode menu under the message box on grok.com and in the app	Auto (picks Fast or Expert) · Fast (default) · Expert (thinks hard) · Heavy (a team of AI agents)	Free accounts get Fast. Expert and Heavy take longer and draw more from your weekly usage pool.	^1,2,3
Claude	Model menu next to the send button; click it, then Effort	Pick a model (Haiku, Sonnet, Opus, Fable on Max and Ultra), then Effort: Low · Medium · High · Extra high · Max (levels vary by model)	Each model marks one level as Default. Low and Medium stretch your limits; High is the best balance; Extra high and Max are for long, hard tasks and use your limits faster.	^4,5,6
Gemini	Tap the model name inside the text box at the bottom (you must be signed in)	Auto (default) or pick Flash-Lite · Flash · Pro, then a thinking level: Standard · Extended · Deep Think (AI Ultra only)	Google says more advanced models and higher thinking levels use more of your limit. On paid plans, limits refresh every 5 hours up to a weekly cap.	^7,8,9,10
ChatGPT	Model menu at the top of the chat; choose GPT-6, then move the Thinking slider	Instant (fast) · Medium · High · Extra High · Pro. Free and Go get Instant plus a Think button	Plus includes Medium and High; Pro adds Extra High and Pro. If you hit a Thinking limit, ChatGPT may continue with another model.	^11,12,13
Copilot	Mode menu at the bottom left of the chat box	Smart (default, decides how hard to think) · Think Deeper · Quick response · Study and learn · Search	Smart is the safe default. Think Deeper is slower and more thorough; Quick response is fastest.	^14,15,16
Perplexity	Model selector at the bottom right of the prompt box	Best (automatic, default) · paid plans: Custom → choose a model and turn Thinking on	Deep Research picks its own models. Perplexity suggests Best for everyday questions and Thinking for complex, multi-step ones.	^17,18,19

Open a chat

Find the model name near the message box (or a mode menu)

Choose a model, then a thinking level

Where the picker usually is (our diagram, not a screenshot)

Newer or older model?

Should I pick a newer or older model?

Usually pick the newest. New models are normally better and often the default. But:

Apps sometimes keep older models in the menu. OpenAI, for example, says its GPT-6 rollout "does not retire older selectable models" (OpenAI¹¹).
If a new model seems worse for something you do often, try the same question on the older one and keep whichever works better for you.
Older or smaller models can be faster and use less of your limit.

Want to know which models are actually best? See our model rankings in plain English and the advanced model guide.

Keep going

AI crash course: start here

What AI chatbots are, how to ask, and how to stay safe. About 10 minutes.

Should I pay for AI?

What free plans really give you, what a paid plan adds, and who needs one.

Sources (19)

Sources checked Oct 22 to Oct 26, 2026.