ChatGPT vs Claude: Similarities and Differences (plus alternatives)

So, you’re getting deeper into using AI tools. But before you go much further, you want to know which company you should be getting into bed with.

Much like Android and iOS, there are loads of things in common, but a few distinguishing differences. Switching teams can be time-consuming as you learn the ins and outs of the other, so it helps to pick a side and stick with it for a while.

Below is an exhaustive (I hope) analysis of what ChatGPT and Claude — and their companies, OpenAI and Anthropic — have that distinguishes each one, what they have in common, and whether there are realistic alternatives.

I use both with paid plans and switch their priority every now and then, so I’m personally aware of their pros and cons. It’s impossible to cover everything, but this should be enough for most everyday users.

The short version

Below is a summary of why you might choose one or the other. I’ve put the broad product differences here first; the model names and limits need more explanation.

If I had to recommend one $20 plan to a general user, I’d still start with ChatGPT Plus. Image generation and the wider voice tools are hard to ignore if you want one subscription to do lots of different jobs.

If your week is mostly writing, reading long files, building small tools or working through code, Claude Pro may make more sense. You can try both for a month without choosing a side for life. (And no, I don’t get a commission from either company for saying that.)

What ChatGPT and Claude have in common

In a normal chat window, the products are much more alike than their fans sometimes admit. Both can:

• Answer questions, edit writing, translate and brainstorm.
• Reason for longer before answering when you select a thinking mode.
• Search the web and cite sources.
• Run a longer research job across the web, files and connected services.
• Read PDFs, spreadsheets, screenshots, photographs, diagrams and charts.
• Write and run code, analyse data and create downloadable files.
• Remember information across chats, subject to your settings and plan.
• Organise work into projects with instructions and reference files.
• Talk by voice on supported devices.
• Connect to third-party services and use tools rather than only producing text.
• Hand longer jobs to an agent that can work through several steps.
• Provide developer APIs, team plans and enterprise controls.

Both also make mistakes. A polished answer is not proof that the answer is correct. Web citations help, but a model can cite a source that does not quite support its claim. For legal, medical, financial or business-critical work, read the original source.

Which models are current?

The model name in your ChatGPT window is only part of the story. OpenAI’s current guide says paid ChatGPT chat uses GPT-5.6 Sol, while Free and Go use GPT-5.6 Luna. GPT-6 Astra is available in Work and Codex on Plus, and as GPT-6 Pro in ChatGPT on eligible Pro and business plans. GPT-6.1 Sol is rolling out in Work and Codex and is also in the API. So if someone shows you a score for GPT-6.1 Sol, don’t assume that is the model answering your ordinary chat.

Anthropic has Opus 5.5, Sonnet 5.5 and Fable 5.1 at the high end. Sonnet is the everyday choice for many paid users. Opus takes on the harder jobs. Fable can run ambitious, long tasks, but its credit rules mean that “available on my plan” does not always mean “included in my monthly price”. Anthropic has announced Haiku 5.5; it was not out when I updated this.

What independent tests say

I looked at Artificial Analysis’s Intelligence Index v4.3.2 because it runs the models through the same ten evaluations. That gives us a better starting point than comparing two companies’ chosen launch slides. It still isn’t a verdict on your own work: change the task, tools or thinking setting and the order can change.

On coding tests, the lead moves around. Artificial Analysis reports Opus 5.5 at 59.6% on Terminal-Bench 4.0 and Sonnet 5.5 at about 64%, against about 60% for Astra. Its GPT-6.1 Sol test found a $0.72 average API cost per benchmark task, compared with $3.26 for Astra. That says something about the test’s cost and setup; it doesn’t tell you what your subscription will charge for a particular coding session.

I also read a Tom’s Guide comparison of everyday prompts. The reviewer liked Claude more for several writing and messy-planning jobs, and GPT for concise answers and tidy plans. I recognise that split from using both, but it was a small set of prompts. If you are choosing a writing assistant, hand both the same awkward draft and see which edit you would actually keep.

Everyday chat, speed and tone

Speed

There is no permanent winner on speed. A fast model such as GPT-5.6 Luna will usually answer before a flagship model that is allowed to think. Claude Haiku, when offered for a task, fills a similar fast-model role on Anthropic’s side. Sonnet sits nearer the middle; Opus and Sol at high reasoning settings can take longer.

The job matters more than the logo. Ordinary chat normally takes seconds. A research report, a large coding change or a computer-use task can take minutes. Server load, tool calls and a site’s own response time can dominate the model’s speed.

Tone

A widespread observation is that Claude’s default prose is often measured, complete and slightly formal. ChatGPT’s default can feel more conversational and can vary more when you change its personality, memory or instructions. That is an observation about defaults, not a testable rule about which model “writes better”.

Both will follow a clear style brief. Give either one a sample, say who the reader is, ban phrases you dislike and ask it to edit its first draft. The gap gets much smaller. Tone also changes after model updates, so I would not make it the sole reason to lock myself into one company.

Coding, remote control and work

Which is better for coding?

The old “Claude has a coding tool and ChatGPT has a chatbot” comparison has expired. Claude Code and Codex can both read a repository, change files, run tests and explain a patch. Claude Code feels at home in the terminal and desktop app. Codex can work locally or in a cloud environment that keeps going when you close your laptop. The recent coding tests give Claude a lead on some hard tasks; GPT-6.1 Sol offers a cheaper model option close to Astra’s broader score. Neither benchmark has seen your project’s odd build script.

I would give both agents the same small issue in a real codebase and read the diffs. Did they run the tests? Did they change anything you didn’t ask them to change? How much tidying did you do afterwards? Those questions tell me more than a one-point lead on a chart.

Remote work: three different meanings

“Remote” covers three quite different arrangements. If your phone is controlling a session on your Mac, the Mac has to stay on. If the job runs in the cloud, you can shut the laptop. An always-on agent can keep starting work while you do something else, which is handy until it starts using a tool or a paid allowance you forgot it had. I would check its permissions and activity rather than treating it like a background tab.

ChatGPT Work, Claude Cowork and Dots

ChatGPT Work is OpenAI’s general work mode for files, apps and a computer. Codex is the coding counterpart, and they share a usage pool on supported plans. Anthropic folded Cowork into the main Claude app in September, with document, slide and design work in chat. Both can make a finished file, but I would open it before sending it to anyone. A good-looking preview can hide a wrong number or a mangled page.

Dots are OpenAI’s next step: an always-on Astra agent gets its own cloud computer and approved tools. At this update, eligible Pro $100, $200 and $500 users aged 18 or over can use them outside the EEA, UK and Switzerland. Business Premium has access worldwide; Enterprise is in admin-controlled beta. Talking to a Dot does not count as a plan message, but a Work or Codex task it starts does. The first month has expanded usage, so I would be careful about judging an ordinary month from a new account’s first week.

Claude has its own ways to keep work moving. Scheduled Cowork tasks can run remotely, and Claude Tag can follow up in Slack for eligible teams. If you want a personal agent living on a cloud computer, Dots are the more direct match. If your work already lives in documents, code and team chat, compare Cowork, Claude Code and Tag together.

Maps and travel planning

ChatGPT Maps puts an interactive map in the conversation. I can ask about an area and then look at the places instead of trying to picture a list of names. It is still worth checking opening hours, routes and bookings in a live map or on the venue’s own site. I would not let a chat answer send me across town to a restaurant that shut last year.

Claude has maps too, but the route to them depends more on where you use Claude. Its iOS app can hand places to Apple Maps, Android can use Google Maps, and Team and Enterprise interactive results can show Google Maps. ChatGPT’s map inside chat is easier to explain and use across a general conversation. Neither is replacing a navigation app.

Images, documents and visual work

If you want an illustration, product mock-up or revised photo, ChatGPT Images 2.5 is the obvious difference. You can give it a prompt or a reference image, then ask it to change part of the result. I still inspect the output closely: a request to fix the lighting can also change a face, a logo or a line of text.

Claude does not make raster photos or illustrations. It can write SVG or HTML for diagrams, charts and interfaces, which is a different job and can give you something easier to edit. Both can read still images and documents. ChatGPT’s input guide includes some video attachments; Claude’s published upload list covers documents and still images rather than a general audio or video upload. I would check a small detail against the original either way.

That makes my choice simple for blog covers or a photograph: ChatGPT. For a chart, a small web page or a document that I will keep editing, I would try both and inspect the actual file. A nice preview does not tell me whether the labels or figures survived export.

Voice, audio and video

Speaking and listening

Both let you talk instead of type, but ChatGPT has more ways to use live media. Its voice modes speak and listen, and eligible mobile modes can take camera or screen input. ChatGPT Record can transcribe and summarise a meeting on macOS for eligible paid plans. I would still check a transcript before repeating a name, date or decision.

Claude voice also handles a two-way conversation across its apps and can use connected tools. Fable is not available in voice. If you want to talk through a question, both can do it. If you want a meeting transcript or live camera input, ChatGPT has the wider set. If you want a finished soundtrack or a controlled voiceover, I would look at an audio service built for that job.

Video input and video generation

ChatGPT can inspect some uploaded videos, but I would not trust it to catch every movement or the exact timing of an event. It may only see parts of the clip. Claude does not list a general video upload workflow in its current file guide. For anything important, I would keep the original footage and timecodes beside the AI summary.

This part changed more than I expected. OpenAI closed Sora’s app and web product on 26 April 2026 and ended its API on 24 September. Claude does not make general videos either. So the old “ChatGPT wins because it has Sora” line no longer holds. If video generation is the reason you are shopping, I would compare current dedicated video services instead.

Search, research and connected tools

Both have quick web search and a slower research mode. A quick search is for a fact that might have changed this morning. Research is for “read these pages, compare them and give me a cited answer”. Either can still misread a source or carry an old date forward. I ask for the page behind a number and open it myself.

The menu names can make this feel harder than it is. ChatGPT has custom GPTs, projects, memory, plugins and connected apps. Claude has projects, styles, memory, Skills and connectors. Both can reach outside services, but the exact app and permissions depend on your plan, region and workplace settings. Check the one connection you need rather than choosing from a marketing screenshot. I explain the wider idea in my guide to running a business with AI and MCP. I also have a Cover Artist example for people who mainly want AI blog art.

Plans and prices

Here are the advertised US prices before local tax, from OpenAI and Anthropic on 1 October 2026. They change plans often enough that I would still check the checkout page. Also, paying for chat does not automatically buy you API tokens for a program you build.

I would not start at $100 or $200 because a benchmark says the top model is better. Try a $20 month first. If the work you actually do keeps hitting a cap, then price the upgrade against the time it saves you. Fable and API credits can add to the bill, so include them rather than comparing only the subscription labels.

How often do people hit the limits?

How often does someone paying $100 or $200 hit the wall? I looked for a representative figure and neither company publishes one for each expensive model. People report it in reviews and forums, especially after long coding sessions, but those posts cannot tell us what share of all subscribers run out. There are also two clocks to watch: the five-hour window and the weekly cap. You can get through one busy afternoon and still find the high-end model unavailable later in the week.

The length of the job helps explain why. In its maximum-effort tests, Artificial Analysis saw Sonnet 5.5 use about 193,000 output tokens per benchmark task, roughly seven times Astra’s output in that test. Opus 5.5 used about 119,000. Those API measurements do not turn neatly into subscription messages, but they show why a long agent job can eat far more of an allowance than a short chat.

For a couple of normal working weeks, I would note each time the app stops me, switches models or asks me to wait. Was I coding, researching or chatting? Was it a large file? Which meter ran out? One awkward project may be cheaper to handle with extra credits. If the same cap interrupts your paid work every week, a higher tier starts to have a clearer case.

Privacy, reliability and connected apps

Connections are where I pay the most attention to permissions. Giving an agent access to email, code or a computer is different from asking it to explain a paragraph. I check what it can read, what it can change and whether I will see an approval step. A citation is also only the beginning of a fact check; the source may not say what the summary claims.

There is a meaningful privacy difference on personal accounts. OpenAI may use your content to improve models unless you opt out under its rules. Anthropic asks consumers to opt in, with separate rules for safety review and feedback. Business and API terms differ again. I would read the current controls before pasting in client work or personal records.

Both can sound certain while getting a date, citation, line of code or detail in an image wrong. I run the tests for code, open the sources for research and inspect the finished file for documents and media. The more tools an agent can use, the more consequential a mistake can become.

So which would I choose?

For a general user paying for one service, I would start with ChatGPT Plus. Native image generation, the wider voice tools, Maps, Work and Codex give you a lot in one $20 plan. I would move up to Pro only after the work I actually do keeps running into a limit. Dots are tempting if you want an agent to keep a job moving, but I would check the region rules and the usage meter first.

I would choose Claude Pro instead if my week centred on long documents, writing I need to polish, Artifacts or terminal coding. I would accept that chat, Code and Cowork share an allowance, and I would price Fable’s credits before setting it loose on a big job. I would keep another service available for images, as annoying as that is.

And I would keep a free account with the other provider. Switching is time-consuming once you have projects, memories, plugins and habits, but it is not a marriage. One paid home and one free escape hatch is a sensible setup. (Some free, though questionable, relationship advice for you, too!) If video generation is the main reason you are shopping, neither of these subscriptions currently solves it.

Alternatives to ChatGPT and Claude

If you already live in Google Workspace or Microsoft 365, I would price Gemini or Copilot before buying a pile of separate AI seats. The connection to the documents people actually use may matter more than a small difference between model scores. Perplexity is worth trying for sourced web research. For image, audio or video production, compare a dedicated tool with the general assistant on the specific file you need to make.

This comparison will probably age within a week. Things change that quickly, so I’ll keep the update box at the top current as the models, tools and limits move.