I have used both of these as my main tool at different points and I pay for one of them now. I have also rewritten this comparison, because it was wrong about three specific things, and all three errors ran the same direction: it said Claude could not do something Claude can do.
Getting that on the record first, because those three claims were the ones doing the deciding.
What this comparison got wrong
Voice. The post said Claude offered "basic voice input on mobile but not a full conversational voice mode with interruption support." Wrong. Voice mode is in beta on every Claude plan including Free, on mobile, desktop, and web. It has a hands-free mode where Claude listens continuously and replies at natural pauses, push to talk for noisy places, and it can use connected tools like Gmail and Google Calendar mid conversation.
Running code. The post said "Claude can analyze and write code but cannot execute it the same way in the standard interface," and handed data work to ChatGPT on that basis. Claude has run code inside claude.ai since the analysis tool shipped on October 24, 2024, and Anthropic noted on November 5, 2025 that it is being replaced by broader code execution that also generates downloadable files and charts.
Memory. The post treated cross-chat memory as a ChatGPT-only advantage. Anthropic lists memory on its plans including the free tier.
Three claims, three tie-breakers, all expired. If you have been choosing ChatGPT for any of them, the reason is gone.
The bigger change: both turned into suites
This is what most comparisons still miss, mine included until today. Neither of these is a chatbot you compare feature by feature. They are both a set of products sharing one subscription, and they have converged on the same shape.
| Anthropic | OpenAI | |
|---|---|---|
| Chat | Claude | ChatGPT |
| Coding agent | Claude Code | Codex |
| Delegated work agent | Cowork | ChatGPT Work |
| Visual output | Claude Design | ChatGPT Images 2.0 |
The two work agents are the striking part. Cowork came to web and mobile on July 7, 2026: you hand it a task, it works across your files, calendar, email, messaging apps, and connected tools, and it escalates to you when it hits a decision only you can make. Scheduled tasks run with no device online because the sessions live on Anthropic's servers.
ChatGPT Work does the same job with the same manners. It takes action across your apps and files, stays on a project for hours by breaking it into steps, and produces finished sheets, slides, docs, and web apps. It has a plan mode where it gathers context, asks questions, and proposes a step by step plan for you to approve before it starts. On desktop it can use your local files and apps.
Both also let the desktop app reach your local files while the web and mobile versions are more limited. Both sync sessions to the cloud. Independently, two companies arrived at the same design, including the ask-before-you-act part.
One caveat on the OpenAI side. Web and mobile access to Work is rolling out to paid plans except Free and Go, while the desktop app carries Chat, Work, and Codex on every plan including Free. So which surface you use changes what you get.
What still actually separates them
Four things, and only four that I would make a decision on.
Writing. Claude is better, and this is the reason I pay for it. Not by a landslide on short work. On anything long, where a brief has several constraints and the piece has to hold voice for two thousand words, Claude drifts less and I edit less. That is a judgement from daily use rather than a benchmark, and you should test it on your own writing rather than take my word.
Raster image generation. ChatGPT does it, Claude does not. Claude produces SVG, charts, diagrams, interactive artifacts, and through Claude Design prototypes and slides that export to PPTX and PDF. That covers more than people expect. It does not cover a photograph, a logo, or a thumbnail. If images are in your workflow, this is a real point for ChatGPT.
One thing to know if you have been picking ChatGPT for images: the thing being retired is the DALL-E name, not the capability. OpenAI set August 30, 2026 as the DALL-E GPT's retirement date, and generation now runs on ChatGPT Images 2.0 on all plans including Free. I went through what to do about that in Midjourney vs DALL-E 3, including exporting anything you want to keep.
Ecosystem size. ChatGPT's third party ecosystem is larger and it is not close. Claude has MCP, which is the better standard and is genuinely broadly supported, plus a plugin marketplace inside Cowork. But if you rely on community built tools, ChatGPT has more of them.
The pricing ladder. This is now the most practical difference, and it gets the least coverage.
Pricing is where the real difference moved
Both charge $20 for the main individual plan, which is why comparisons stop there. The ladders on either side of that $20 are not the same shape at all.
ChatGPT runs Free, Go at $8, Plus at $20, Pro at $100 and $200, then Business and Enterprise. Claude runs Free, Pro at $20 or $17 if you pay for the year, then Max at roughly five times and twenty times Pro usage, starting around $100.
Two consequences worth having in mind.
ChatGPT has a cheap tier and Claude does not. If your use is regular but light, $8 exists on one side and the jump is straight to $20 on the other. That matters to more readers than any capability on the chart above.
Claude's top model is not included at $20. Fable is not on the free tier at all, on Pro it runs on pay as you go usage credits rather than your plan allowance, and only on Max is it included, up to half your weekly limit. So Pro buys you the option to buy Anthropic's best model, not the model.
Both now sell usage credits when you run out, and both meter some models more heavily than others. Claude's limits work on a rolling five hour window with weekly caps on paid plans, and Max carries two weekly caps, one across all models and one for Sonnet alone. I cannot forecast a month on either product without doing arithmetic, and I think that is a fair criticism of both.
The free tiers, which is where most people should start
Both free tiers improved enough in the last year that the advice "just pay for one" has stopped being obviously right, and almost no comparison has caught up.
ChatGPT free now includes image generation, voice, memory, and web search. Not a trial, not a teaser. If you open it twice a week to draft something or check a fact, that is a working product. What you are missing is headroom: hit the cap and you get moved to a lighter model, which you notice most when it happens halfway through something.
Claude free includes voice mode, code execution, memory, connectors, web search, and access to Claude Code, Cowork, and Design within limits. What it excludes is Fable, the higher allowance, and the priority queue when Anthropic is busy.
The honest test is not a feature comparison, it is a fortnight. Use whichever you are drawn to, and pay attention to two things: how often you hit a ceiling, and whether you hit it while doing something that mattered. Never hitting one means stay free. Hitting one occasionally and mildly resenting it means ChatGPT Go at $8. Hitting one mid-task and losing your thread is the $20 signal, and you will recognise it without needing a chart.
I have written up each product on its own terms if you want the detail rather than the comparison: the Claude review and the ChatGPT Plus review. The second one is mostly about which of six tiers to buy, which turned out to be the more useful question than whether $20 is worth it.
The differences that only show up after a month
Feature lists do not capture these, and they are what people actually end up caring about.
Personality. Claude is calmer and less eager to agree with you. ChatGPT is warmer and chattier. For work I prefer the first, because a model that pushes back saves me from my own bad instructions. For casual use the second is more pleasant. This is taste, not quality, and it is worth noticing which one you find yourself arguing with less.
Refusals. Both refuse less than they did two years ago. Claude is still slightly more cautious on personal advice, ChatGPT slightly more cautious on politically sensitive ground. Neither pattern is a reason to choose, unless your work sits exactly on one of those lines, in which case you already know.
Where the lock-in lives. This one is practical. ChatGPT's global memory and custom GPTs accumulate into something personalised, which is useful and also means switching costs you that accumulation. Claude's Projects hold context per piece of work, so the lock-in is per project and easier to unpick or move. If you are the sort of person who might switch tools next year, the second shape is friendlier.
Reliability. ChatGPT takes more traffic and shows more strain during big news cycles. Claude has been steadier through 2026, though it has had outages of its own. Anthropic has also been buying compute aggressively, including an agreement with SpaceX for the whole capacity of the Colossus 1 data centre, described as over 300 megawatts and more than 220,000 GPUs, aimed at Pro and Max subscribers. Whether that shows up as fewer slowdowns is something you feel over months rather than something I can measure.
Speed. Close enough that I have stopped tracking it. Both are fast enough for conversation, and the reasoning modes on both are deliberately slower in exchange for better answers.
If your actual question is about coding tools rather than these two chat products, the comparison you want is different again: top AI coding assistants for beginners covers the editor-based options.
A claim I am retracting
Earlier versions of this post led with Claude's 200,000 token context window as a headline advantage over ChatGPT's 128k on Plus.
I no longer think anyone can make that claim cleanly. Anthropic's pricing page says 200k for Pro. Anthropic's help centre says paid plans on newer models reach up to 1M tokens, with others at 500K or 200K. Those are both first-party pages and they disagree.
The practical guidance is duller and more useful: context depends on which model you pick, both products handle long documents far better than they did, and if you are working at the edge of a window you should test with your actual documents rather than trusting a number in a comparison article.
Task by task, corrected
Long form writing. Claude. The clearest difference between them and the one I would spend money on.
Coding. Claude for following a detailed spec without drifting. The old ChatGPT advantage of running code is gone. For real project work neither chat window is the right tool: use Claude Code or Codex. My workflow is in how to use Claude AI for coding, and if you would rather the assistant lived in your editor, Cursor vs GitHub Copilot covers what those cost once agent usage is metered.
Images. ChatGPT, for raster output. Claude Design covers slides and prototypes if that is what you actually needed.
Voice. Genuinely close now. Both do full spoken conversation with tool access. I use neither much at a desk, and I suspect most people who call voice essential are describing cooking and driving.
Long documents. Both are strong. I am no longer willing to hand this to Claude on a context number, for the reasons above.
Research. Roughly equal, and neither beats a dedicated tool. Perplexity cites sources for every claim, which is a different way of working rather than a better version of the same thing.
Delegated multi-step work. Genuinely equal on paper, and new enough on both sides that I would not make a subscription decision on it yet. Try whichever you already pay for.
Reasoning. Close at the top. Claude shows its thinking, and OpenAI folded its old o-series into thinking levels inside the GPT-5 family, retiring o3 on August 26, 2026 after GPT-4o and others went in February. Being able to read the working is worth more than it sounds, because a confident wrong answer with no trace is expensive.
Who should pick which
Pick Claude if you write for a living, if you code and value a model that holds a specification, or if you want per-project context rather than one global profile. It is my daily tool and the writing quality is why.
Pick ChatGPT if you need real image generation, if the third party ecosystem is what you are buying, or if you want a cheap paid tier because $8 is the right amount to spend.
Start free on either. Both free tiers now include voice, memory, and image or code capability that used to be paid. A week on your own real tasks tells you more than this article can.
Pay for both only if you have a specific reason. A year ago $40 for the pair was the obvious answer for heavy users. Now Claude plus ChatGPT's free tier covers most of it, which is the setup I run.
If Gemini is on your list too, that matchup is in ChatGPT vs Google Gemini, and it matters most if your work already lives in Google Docs and Gmail. For the underlying models rather than the products, see GPT-5.6 vs Claude 5.
My honest take
I use Claude and keep ChatGPT's free tier for images. That has not changed. What changed is that I can no longer justify it with a feature list, because the feature list mostly evened out.
What I would say to someone deciding today is that this stopped being a capability comparison and became a taste and pricing comparison. Do you write enough for Claude's prose advantage to be worth $20. Do you need images enough for that to swing it. Is $8 the right budget rather than $20. Those are answerable questions, and they are more useful than a scoreboard of who has voice.
The uncomfortable lesson from rewriting this is about comparison articles generally, including the ones on this site. Three claims here were true when written and quietly became false, and nothing in the page told a reader they had aged. That is the normal failure mode of this genre. Check the date on anything you read about these two products, including this, and check the vendor's own page before you pay.
Verified on August 11, 2026. Claude's voice mode, memory, code execution, plan structure, and Cowork details come from Anthropic's own support and pricing pages. The ChatGPT tiers, ChatGPT Work behaviour, image generation changes, and model retirement dates come from OpenAI's published pages and announcements as indexed by search, because OpenAI blocks automated access to its own site, so that side is second hand in a way the Anthropic side is not. Both companies ship often. Treat the dated claims as dated.
