Choosing a model and effort.
Match how capable and how thorough the AI is to the job in front of you, then step up only when the answer needs it.
- 6 minute read
- Claude and ChatGPT
- As at October 2026
Two choices.
Every time you use Claude or ChatGPT, two settings shape the answer. You may not have noticed choosing either.
- The model is which AI you’re talking to. Bigger models know more and reason better. Smaller ones are quicker and cheaper to run.
- The effort is how long the model thinks before it answers. More effort means it considers more possibilities and checks more of its own steps. It takes longer doing it.
Different products use different words for effort: thinking, reasoning or effort. Some show a switch, others a menu of levels. They all do the same job.
Neither setting finds facts you didn’t give it. A wrong brief at high effort is still a wrong brief. It just took longer.
Speed, cost and quality.
Every step up buys more depth and costs you something else.
| Setting | Lower | Higher |
|---|---|---|
| Speed | Quicker answers | A longer wait |
| Cost | Less of your plan or bill | More of your plan or bill |
| Depth | Fine for clear, bounded jobs | Better when decisions depend on each other |
On a subscription, the cost is the wait and how quickly you reach your plan’s usage limit. If you pay by use, it’s the bill.
Most everyday business writing doesn’t need the biggest model at the highest effort. Using one anyway mostly makes you wait.
Find your starting point.
Pick the job closest to yours. Each one suggests a starting point in Claude and ChatGPT, plus when to step up. Every example is invented.
A quick email
Reply to a customer who wants to move Thursday’s appointment to Friday.
- In Claude
- Sonnet 5.5 at low effort
- In ChatGPT
- Instant
Why: The facts are in front of it and the finish line is obvious. More thinking just makes you wait for the same email.
Step up when the reply has to handle a complaint or a promise you need exactly right. Try medium.
Summarise a policy
Turn a twelve-page staff leave policy into a one-page summary the team will actually read.
- In Claude
- Sonnet 5.5 at medium effort
- In ChatGPT
- Medium
Why: There are a few decisions inside it: what matters, what to leave out and how to keep the meaning intact.
Step up when people will rely on the summary for decisions, or it keeps missing an exception. Try high and ask it to name the clause behind each point.
A complex analysis
Compare three supplier quotes on price, contract terms and risk, then recommend one.
- In Claude
- Opus 5.5 at high effort
- In ChatGPT
- High
Why: Several decisions pull on each other and a mistake costs real money. This is where extra thinking earns its keep.
Step up when it’s genuinely hard and high still misses things. Try the level above: extra high in Claude, Extra High in ChatGPT. Keep the most capable tier, Fable 5.1 or ChatGPT’s Pro, for the hardest problems.
Build something
Build a simple booking page, or a small tool that tidies a weekly report.
- In Claude
- Opus 5.5 at high or extra high effort, in a coding tool such as Claude Code
- In ChatGPT
- Codex, OpenAI’s coding tool, on a high setting
Why: Building means many connected steps, so it rewards a capable model that plans and checks its own work. Coding tools usually start on a higher setting for this reason.
Step up when a change keeps breaking something else. Before reaching for max, give it a clearer brief and a smaller piece to do.
The tiers at a glance.
Both companies offer a quick option, an everyday one, a deep one and a most capable one. Here’s how they line up as at October 2026.
Claude models
- Haiku 5.5Quickest and lightest. Simple, high-volume jobs.
- Sonnet 5.5The everyday all-rounder. Fast and capable.
- Opus 5.5Complex work with several moving parts.
- Fable 5.1The most capable. The hardest, longest problems.
Effort in Claude runs low, medium, high, extra high and max. Bars show relative pay-by-use prices from Anthropic’s published rates. Which models you see depends on your plan.
ChatGPT thinking levels
- InstantFast answers to everyday questions.
- MediumThinks first. Good for a few moving parts.
- HighThinks longer, for decisions that depend on each other.
- Extra HighThe deepest thinking level.
- ProOpenAI’s most thorough option, on the top plans.
In ChatGPT these levels belong to GPT-6, OpenAI’s current model on paid plans. Free plans use a lighter model with fewer choices. Pro has usage limits.
If a name here has moved on by the time you read this, look for its place in the line. A newer quick model still does the quick jobs.
Start lower. Step up when it’s needed.
Start one level lower than you think you need. Step up when:
- the answer misses something a careful person would catch
- several decisions depend on each other
- a mistake would cost money or trust
- it keeps getting the same thing wrong after you’ve improved the brief
Step down when a higher setting gives you the same answer, just slower.
The honest test is to give two settings the same brief and compare what comes back. If the higher one didn’t improve on the lower one, you’ve learned something useful about that job.
Before you answer, list the assumptions you’re making and anything in my brief that conflicts. Then give me your answer and mark anything you’re unsure of.
How I use it.
Medium is my everyday setting. Most of my writing and summarising happens there and I rarely see a difference going higher.
I step up to high when decisions depend on each other, or when a mistake would cost real money or trust. I keep the very top settings for genuinely hard problems, like building a new tool.
We use Claude every day at The Gallery Real Estate, where this way of working was proven. Whatever the setting, a person still reads the answer, checks it against the source and decides whether it goes.
I’ve written more about both questions on my own site.
Common questions.
Should I pay for the most capable model?
Only if your work needs it. A paid plan on one tool, used well at everyday settings, covers most of what a business does. The top tier earns its cost when you regularly do long, complex work.
Is Claude or ChatGPT better?
Both are good. How you set it up matters more than which you choose. Pick one for the whole team so everyone shares what works.
Does more effort make the answer more accurate?
Not always. It helps with work that needs reasoning. It can’t fix a brief that’s missing facts and a slower answer can still be wrong. It’s often more convincing, which is why a person still checks it.
Why do the same settings give different answers in different tools?
Each model is trained differently and each is updated often. That’s also why the right choice changes over time. Re-test your common jobs every few months.
I can’t find these settings. Where are they?
They move as products update and they vary by plan. Look near where you type for the model’s name or a thinking or effort option. If you can’t see one, your plan may be choosing for you.
What if I pay by use?
Then the price of each answer matters more. As at October 2026, Anthropic’s published rates put Opus 5.5 at twice the price of Sonnet 5.5 and Fable 5.1 at two and a half times Opus 5.5. Haiku 5.5 costs a small fraction of Sonnet 5.5. Higher effort also uses more of whichever model you pick.
What to do next.
Pick one job you do every week. Run it at your usual setting, then one level lower. If the answer is just as good, you’ve saved yourself time on every run from now on.
This guide is as at October 2026. The names will change. The habit of starting lower and stepping up won’t.
Property manager? Sidekick offers coaching and the AI build for property management owners. Thomas is the AI Partner there.
Get the next guide.
Leave your name and email and we’ll send you the next one when it’s ready. One email, no newsletter.
You’re on the list.
We’ll email you when the next guide is ready. Nothing else.