Which one should you actually use
I run my business on Claude and I sell Claude guides. Read the rest of this page with that in mind — and then notice that I am about to hand you three places it loses, two corrections to my own claims, and a prompt that might tell you to cancel. A comparison you cannot trust is worth nothing to either of us.
Everything here was re-checked against the vendors' own pages on 08/09/2026. This space moves weekly, so where something changed recently, I have dated it.
Reach for ChatGPT when
- You need an actual image made — a picture, an illustration, a mockup
- You want something to fire when an email or a message arrives, set up in one place
- You want a quick answer and depth is not the point
- You are already deep in its memory of you and moving would cost more than it saves
Reach for Claude when
- The job is long and structured — a document, a plan, a build with many parts
- You need it to hold a big pile of context at once (the numbers are further down and they are not close)
- You are working on the same project across weeks, not minutes
- You want it working in your real files rather than in a chat box
That is the short version. Below are two prompts that beat any generic comparison, including this one, because they use your work rather than someone else's.
I pay for one AI subscription and I want to know if it is the right one. Do not be diplomatic. Pick one. First, ask me these questions one at a time and wait for each answer: 1. What did I actually use you for in the last month? If you can see our history, go and look rather than asking me to remember. 2. Of those jobs, which ones took more than twenty minutes of back-and-forth? 3. Do I ever need an actual image made - a picture, an illustration, a mockup? 4. Do I want things to run on a schedule or when an email or message arrives? 5. Do I work in files and folders on my computer, or only in a chat window? 6. What am I paying per month right now, and for which tool? Then give me a straight answer in this shape: A. Which of ChatGPT and Claude my actual work suits better, and the single biggest reason. B. The jobs from my list that the OTHER one would do better, and whether that is worth a second subscription or whether the free tier covers it. C. One thing I am currently paying for and not using. D. What I would lose by switching. Be specific, not reassuring. If the honest answer is that either would be fine and I should just pick the cheaper one, say that.
I want to run a fair head-to-head between ChatGPT and Claude on MY actual work, rather than trusting anyone's comparison article. Help me design it so the result means something. The job I want to test: [REPLACE THIS with one real piece of work you do often. Make it something you can judge properly - a client email, a plan, a summary of a long document, a piece of code, a first draft.] Do this: 1. Write me ONE prompt for that job, detailed enough to be a fair test, that I can paste into both tools unchanged. No tool-specific tricks in it. 2. Tell me what to attach or paste alongside it so both get the same information. 3. Give me a scoring sheet with 5 things to judge, written for my job specifically - not generic ones like "accuracy". Things like "did it keep my client's name right the whole way through" or "did it invent a number". 4. Warn me about the two ways I am most likely to rig this test without meaning to. 5. After I paste both answers back to you, score them against the sheet and tell me which won and by how much. If it is close, say it is close rather than picking a winner for the sake of it.
Run the second one before you switch anything
Most people who move tools move because of a video, not because of a test. Twenty minutes running the same real job through both will tell you more than every comparison page on the internet, mine included.
Where Claude actually loses
Three of them. I named these on camera; here they are checked at source, including the one that has shifted under me since I filmed it.
Images. Claude does not make them.
Anthropic says it plainly in their own help centre: Claude does not generate photos or illustrations the way image tools do. It will build you a diagram, a chart or an interactive thing in code — genuinely good ones — but if you want a picture, you are opening something else. ChatGPT generates images on every paid tier, and Pro gets them unlimited and faster.
This one is not close and it is not changing this week. You can wire an image model into Claude yourself through a connector and it works — I do it — but that is a job, not a feature, and Anthropic does not document it as a supported path. So I am not scoring it as a tie.
Scheduling. ChatGPT keeps it in one place; Claude has four.
This is the one that annoys me daily. In ChatGPT you set up a scheduled task in the app and that is where scheduled tasks live. In Claude, depending on what you want, it is: Cowork scheduled tasks (time-based, all paid plans), Claude Code routines in the cloud, desktop scheduled tasks running locally, or a /loop inside a session. Four places, four sets of rules, and no single screen showing you everything you have running.
Claude also gates parts of it by plan, and ChatGPT caps how many active tasks you get by plan — 5 on Plus, 15 on Pro. Both have limits. Only one of them has four front doors.
The in-app browser — but this one has nearly closed
When I filmed, ChatGPT's in-app browsing was the clearer win and it still feels quicker to me in the desktop app. Two things happened in the last month that I have to put on the record:
09/08/2026 — OpenAI retired Atlas, its standalone browser, and folded the agentic browsing into ChatGPT and Codex instead. 26/08/2026 — Anthropic shipped a browser built into Claude Cowork on the desktop app, in beta for Pro, Max and Team: it opens in a side panel and Claude navigates, reads, clicks and types.
So neither of them sells a browser any more, and both drive one for you. "Claude cannot browse" would be flat wrong — it has web search on every plan including free, a Chrome extension on every paid plan, and now a browser in the side panel. Call this a preference, not a gap.
A correction, in the direction that does not flatter me
I said ChatGPT has trigger-based runs and Claude does not. That is not right in either direction. ChatGPT's scheduled tasks do fire on Gmail, Slack and GitHub events on Plus and up, not just on timers — so that half stands. But Claude Code routines can be triggered by an API call or a GitHub event too. And ChatGPT is not perfectly unified either: Codex runs its own automations, even though they now show up in the same Scheduled view.
The honest version is not "ChatGPT has triggers and Claude does not". It is "ChatGPT has one place to look and Claude has four". Which is still a real loss, just a different one.
Where Claude wins
Same treatment. And the third one is where I have to correct myself hardest.
Claude Pro — $20/mo
- 1,000,000 tokens on the newest models
- 500,000 on the mid-tier models
- 200,000 on the older ones
- Summarises earlier messages so a conversation can run past the limit
ChatGPT Plus — $20/mo
- 256,000 tokens on reasoning
- 54,000 on the fast instant mode
- Pro ($100+) lifts those to 400,000 and 128,000
- Free sits at 27,000 on instant
That is the single biggest verifiable difference between the two at the same price. It is not a rounding error — it is the gap between pasting in three chapters and pasting in the whole book. (Small honesty note: Claude's own pricing page cards still say 200K. The help centre carries the current figures, and that is what I have used.)
Long, structured work that has to stay consistent
Context is not a spec-sheet number, it is what stops page 40 contradicting page 2. If your work is one long connected thing — a plan, a document, a build with a dozen dependent parts — the tool holding four times as much of it will hold your thread better. This is the reason I stayed.
Living inside one project for weeks
Both have Projects. Claude's also do retrieval over the files you upload, so it pulls the relevant bit out of a big pile rather than needing the whole pile in front of it. For ongoing client work, that is the difference between re-briefing it every Monday and just carrying on.
Working in your actual files — and here is my correction
I say on camera that Claude Code has no real equivalent. It does, and I was wrong. OpenAI ships Codex — a terminal CLI, an IDE extension for VS Code, Cursor, Windsurf and JetBrains, a cloud agent, and an iOS app — and it is included in ChatGPT Plus, exactly the way Claude Code is included in Claude Pro.
So the honest version: both $20 tiers now include a terminal agent that works in your real files. If you have been choosing Claude because the other one had nothing like it, that reason has expired. Choose on which one you actually like driving. I still prefer Claude Code, and that is a preference, not a fact.
What I checked, and where
Every number on this page came from the vendors themselves on 08/09/2026: OpenAI's pricing page and help centre for ChatGPT plans, tasks, memory and export; the Codex pricing and product pages for what Plus includes; Anthropic's pricing page, the paid-plan context-window article, the Cowork browser announcement, the Cowork and Claude Code scheduling docs, and the memory and data-export articles.
Two things I could not verify from an official source, so they are not stated as fact anywhere above: Australian dollar pricing (neither company publishes a country price table) and the launch date of Claude Code routines (the doc carries no date).
How to run both without paying twice
The two ladders have converged to the point of being funny. Both run free, then $20, then $100, then $200 a month in US dollars. ChatGPT has an $8 Go tier underneath that; Claude has nothing at that price. Claude Pro drops to $17 a month if you pay for the year up front.
So a second subscription is another $20 USD a month. Expect somewhere north of A$30 once GST and the exchange rate land — neither company publishes an Australian price list, so check your own checkout rather than trusting a number off a blog.
Pay for the one you live in. Use the other's free tier.
This is the answer for almost everyone. ChatGPT's free tier still gives you image generation (limited), built-in browsing and a few scheduled tasks — which covers the three things Claude loses at. Claude's free tier includes web search. Whichever you pay for, the other's free account handles its own wins.
Do not reach for the API to save money
No Claude subscription includes API usage — Anthropic states that outright for Pro. OpenAI's API is prepaid, minimum $5, credits expire after a year and auto-reload is on by default. For a non-developer it is a worse deal than a second $20 subscription, not a better one: no memory, no Projects, no interface, and a bill that moves.
If you are hitting limits, buy credits inside the product
Both sell in-product overage now. Topping up the tool you already use beats adding a whole second subscription you will use twice a month.
Annual on one, monthly on the other
Commit for the year on the one you open every day and take the discount. Stay monthly on the other so you can drop it the week you notice you have not touched it.
Moving between them
Your chat history does not move. Neither direction.
Both let you export. Neither will import the other's conversations. Anthropic goes further and says exported Claude data cannot even be imported back into another Claude account. Nobody is building this. Plan around it rather than waiting for it.
Which sounds worse than it is, because the thing worth moving is not your old chats. It is what the tool has learnt about you.
Memory moves — but only one way, into Claude
Claude has a proper import flow for memory from other providers, ChatGPT included: you run an export prompt in the old tool and paste the result into Claude's import screen. Free, Pro, Max and Team, on web and desktop. Anthropic flags it as experimental and still in active development, so treat it as a head start rather than a clean transfer. There is no equivalent going the other way.
Getting your data out of ChatGPT
Settings, or privacy.openai.com. Available on Free, Go, Plus and Pro. Not self-service on Business or Enterprise — there it is a workspace owner job.
Getting your data out of Claude
Settings, then Privacy, then Export data. Web or desktop only — you cannot do it from the phone apps. The download link expires after 24 hours, so start it when you are actually near your computer.
Turning memory off, in either
ChatGPT: Settings, then Personalisation, then Memory. You get an editable summary you can correct by typing into it — with OpenAI's own caveat that the summary will not include everything it remembers, and that fully clearing it means clearing chats, archived chats, files and connected apps too. Claude: Settings, then Memory. You can view individual topics, edit them, delete them, reset the lot or pause it. On Team and Enterprise, Claude's memory is off unless an owner switches it on.
One dated note
Anthropic set 09/09/2026 as the cut-off for exporting legacy memory. If you are reading this before then and you have years of it in there, that is a today job. If you are reading this after, it has gone and there is nothing to do about it.
The bit I actually want you to take away
I sell Claude guides and I still think you should run the test in prompt 2 rather than take my word for it. If ChatGPT wins on your work, use ChatGPT. The tool is not the point — the thing you are building with it is. I would rather you trusted the next thing I tell you.
Ready to go deeper?
Picking the tool is the easy half. The useful half is building the thing you picked it for.
Wright Mode Membership
Join a community of women using AI to take real work off their plate. Live calls, copy-paste prompts and workflows, and ongoing support.
Claude Masterclass
Learn to use Claude like a pro — from prompting fundamentals to building real workflows that save hours every week.
Claude Code Masterclass
Go beyond the chat box. Build automations, process data and create tools with Claude Code — no developer experience needed.