ChatGPT vs Claude vs Gemini

Three assistants, one subscription — which one should you actually pay for?

The verdict

ChatGPT scores highest: 90/100 on our everyday chat tasks

ChatGPT 90, Claude 87, Gemini 84 out of 100, a 3-point gap to Claude. Every answer and every judge's reasoning is downloadable below.

Ranked on our tested everyday chat score (18 identical tasks, three-lab blind panel, 10 Aug 2026)

What each one looks like

Gemini homepage
Gemini as it appeared on 13 Aug 2026 · gemini.google.com

We have no dated capture of ChatGPT or Claude yet, so none is shown.

Side by side

Every cell is either a fact with the source and the date we read it, or a sentence saying we hold no verified answer. There is no third option: nothing here is filled in from memory, and a row only appears when at least one of these tools has something real in it.

Swipe the table sideways for the other columns.

MeasureChatGPTClaudeGemini
Our tested score — everyday chatIdentical tasks, marked blind by three judges from three rival AI labs. Receipts on every score.90/10018 tasks · 10 Aug 2026receipts87/10018 tasks · 10 Aug 2026receipts84/10018 tasks · 10 Aug 2026receipts
Our tested score — writingIdentical tasks, marked blind by three judges from three rival AI labs. Receipts on every score.90/10018 tasks · 9 Aug 2026receipts87/10018 tasks · 9 Aug 2026receipts87/10018 tasks · 9 Aug 2026receipts
What users rate itApp-store ratings, ranked by rating weighted by review count — so a 4.88 from three million reviews outranks a 4.9 from four thousand.4.78★61.9m reviews on App Store + Google Play · #4 of 14 we can rankApple App Store (iTunes Lookup API) · 17 Aug 20264.54★895k reviews on App Store + Google Play · #14 of 14 we can rankApple App Store (iTunes Lookup API) · 17 Aug 20264.59★43.9m reviews on App Store + Google Play · #13 of 14 we can rankApple App Store (iTunes Lookup API) · 17 Aug 2026
What people are sayingFrom our own 90-day sweep of Reddit, X, YouTube, Hacker News and the web. The count always travels with the word.Negative8 of the 10 posts that took a side (31 scanned)Negative11 of the 13 posts that took a side (116 scanned)Negative16 of the 18 posts that took a side (68 scanned)
Cheapest plan we verifiedRead off the vendor's own pricing page and dated. Where a vendor lists plans it only renders in a browser, we hold no cheapest figure and say so rather than quote the top tier.£7/mogo planchatgpt.com/pricing · 16 Aug 2026$20/mopro plan · US priceclaude.com/pricing · 17 Aug 2026£7.99/moai plus planone.google.com/about/plans · 17 Aug 2026
Free versionConfirmed on the vendor's own pricing page, with the date we read it.Yesconfirmed 16 Aug 2026chatgpt.com/pricing · 16 Aug 2026Yesconfirmed 17 Aug 2026claude.com/pricing · 17 Aug 2026Not confirmed on their pricing pagecheck
Trains on what you typeThe vendor's own published answer. No line renders unless we found the sentence word-for-word in a document we fetched.They use your conversations to train their AI models, but you can change your settings to stop this.openai.comThey use your conversations to train their AI models, but you can turn this off in your account settings.anthropic.comThey use your chats and interactions with Gemini to train and improve their AI models.policies.google.com
How long they keep itThey keep your conversations until you choose to delete them, after which they are removed within 30 days.openai.comThey keep your data for as long as they decide it is necessary for their purposes.anthropic.comThey keep your data until you delete it or set it to be removed automatically.policies.google.com
They warn it can be wrongThe company warns that the tool can make mistakes and says you should check the answers yourself before using them.openai.comThe company warns that its AI can make mistakes and tells users to independently check the accuracy of the results.anthropic.comThe company warns users not to rely on the service for medical, legal, financial, or other professional advice.policies.google.com
Last app update we sawThe date on the vendor's own app-store listing when we last read it — a rough read on whether the product is still moving.7 Aug 2026Apple App Store (iTunes Lookup API) · 17 Aug 202613 Aug 2026Apple App Store (iTunes Lookup API) · 17 Aug 202615 Aug 2026Apple App Store (iTunes Lookup API) · 17 Aug 2026

Head-to-heads we ran between them

A battle is one head-to-head run on one day, on a single category’s tasks; the scores in the table above average a whole suite. The two can legitimately disagree, and both are published.

Each one on its own

ChatGPT

OpenAI

You use it to answer questions, write text, create images, and write computer code.

In their words: “ChatGPT helps you get answers, find inspiration, and be more productive.” — chatgpt.com · read 9 Aug 2026

90/100 everyday chat · 90/100 writing · receipts ↓

  • go£7/moverified 16 Aug 2026
  • plus£20/moverified 16 Aug 2026
  • pro£89/moverified 16 Aug 2026

every plan read on chatgpt.com16 Aug 2026

OpenAI didn’t upgrade ChatGPT. It lobotomized the part that made it worth using.

a negative post from our 90-day sweepread it ↗

full ChatGPT page → · visit chatgpt.com

Claude

Anthropic

It helps you analyze data, write code, and think through your hardest work.

In their words: “Claude is Anthropic's AI, built for problem solvers.” — claude.ai · read 9 Aug 2026

87/100 everyday chat · 87/100 writing · receipts ↓

  • max from$100/moverified 17 Aug 2026
  • pro$20/moverified 17 Aug 2026

every plan read on claude.com17 Aug 2026

YSK: A Whole Bunch of People’s Claude Chats Are Publicly Accessible Online, and There’s Some Wildly Private Stuff in There

a negative post from our 90-day sweepread it ↗

full Claude page → · visit claude.ai

Gemini

Google

It helps you with writing, planning, and brainstorming.

In their words: “Meet Gemini, Google’s AI assistant.” — gemini.google.com · read 9 Aug 2026

84/100 everyday chat · 87/100 writing · receipts ↓

  • ai plus£7.99/moverified 17 Aug 2026
  • ai pro£18.99/moverified 17 Aug 2026

every plan read on one.google.com17 Aug 2026

full Gemini page → · visit gemini.google.com

Questions people ask

ChatGPT or Claude or Gemini — which should I choose?

ChatGPT scores highest: 90/100 on our everyday chat tasks. That is based on our tested everyday chat score (18 identical tasks, three-lab blind panel, 10 Aug 2026). ChatGPT 90, Claude 87, Gemini 84 out of 100, a 3-point gap to Claude. Every answer and every judge's reasoning is downloadable below.

Which of ChatGPT, Claude, Gemini is cheapest?

ChatGPT starts at £7 a month on its go plan; Claude starts at $20 a month on its pro plan; Gemini starts at £7.99 a month on its ai plus plan. Every figure was read off the vendor's own pricing page and is dated on this page.

Do ChatGPT, Claude, Gemini train on what you type?

ChatGPT: They use your conversations to train their AI models, but you can change your settings to stop this. Claude: They use your conversations to train their AI models, but you can turn this off in your account settings. Gemini: They use your chats and interactions with Gemini to train and improve their AI models. Each of those answers is quoted word-for-word from the vendor's own published document, linked on this page with the date we read it.

How do you test these?

ChatGPT scored 90/100 on our everyday chat suite (18 tasks); Claude scored 87/100 on our everyday chat suite (18 tasks); Gemini scored 84/100 on our everyday chat suite (18 tasks). Every tool answers the identical published tasks; three judges from three rival AI labs mark every answer blind against a fixed rubric, and mechanically checkable rules such as word limits are enforced by the harness in code. The raw answers and judge verdicts are downloadable.

No vendor pays for placement, position or inclusion, and the code that produces the verdict above cannot see which outbound links earn us anything. How we make money →

More head-to-heads: Cursor vs GitHub Copilot vs Windsurf · Lovable vs Replit vs v0 · Otter.ai vs Fireflies.ai vs Fathom · Grammarly vs Jasper vs Copy.ai vs Notion AI · Perplexity vs NotebookLM
Every tool head-to-head → · Every tool we track → · Methodology →