ChatGPT vs Claude: Which is better in 2026?

ChatGPT Plus and Claude Pro both cost $20 a month, and team seats cost the same, so choose by the work you do.

Tested hands-on: 4 tests on a MacBook Pro (M1 Pro), October 21, 2026. How we tested

Some links on this page are affiliate links. If you buy through them we may earn a commission at no extra cost to you. This never affects our verdicts.

Quick answer

At a glance

ChatGPT and Claude at a glance
ChatGPT Claude
Free plan Yes Yes
Paid plansFrom $8/moFrom $20/mo
Our testsWon 0 of 43 ties Won 1 of 43 ties
Best forPeople who want one assistant for writing, research, images and voice, with room to move from Free to Go, Plus or a Pro tier as their use grows.Writers, analysts and developers who work with long documents or code and do not need image generation.

ChatGPT, from OpenAI, and Claude, from Anthropic, both answer questions, draft and edit text, read the files you upload and search the web. Each has a free plan with web search, memory and voice mode, apps for the web, desktop, iOS and Android, and a $20 monthly plan. They differ in the details: ChatGPT creates images and has a cheaper $8 Go plan, while Claude lists a context window of up to 1 million tokens and a yearly price on Pro. Both include a coding tool: Codex on every ChatGPT plan, Claude Code on Claude's paid plans. This comparison sets out plans, limits and privacy settings side by side, using each vendor's own pricing and help pages.

Pricing

ChatGPT and Claude plans and prices
PlanMonthlyYearly (12-month total)
ChatGPT
FreeUnlimited text chats subject to abuse guardrails; limited uploads, image creation, voice and deep researchFree Free plan
GoBilled monthly only; may include ads$8/mono yearly price
PlusBilled monthly only$20/mono yearly price
Pro 100Billed monthly only$100/mono yearly price
Pro 200Billed monthly only; more usage than Pro 100$200/mono yearly price
Pro 500Billed monthly only; highest usage, includes Ultrafast$500/mono yearly price
Business (Standard seat)Per user/month; $20 per user/month billed annually; minimum 2 seats$25/mo$240/yr ($20/mo)
Business (Premium seat)Per user/month; $100 per user/month billed annually; 5x Standard usage$125/mo$1,200/yr ($100/mo)
EnterpriseContact salesCustom
Claude
FreeSonnet and Haiku models; up to 5 projectsFree Free plan
Pro$17/month with annual billing ($200 billed up front)$20/mo$200/yr ($16.67/mo)
Max 5xBilled monthly only; includes $100/month in API credits$100/mono yearly price
Max 20xBilled monthly only; includes $200/month in API credits$200/mono yearly price
Team (Standard seat)Per user/month; $20 per user/month billed annually; 2 to 150 seats$25/mo$240/yr ($20/mo)
Team (Premium seat)Per user/month; $100 per user/month billed annually; 5x Standard usage$125/mo$1,200/yr ($100/mo)
EnterprisePer user, billed annually ($20/seat/month), 20-seat minimum, plus usage at API ratesno monthly price$240/yr, billed yearly ($20/mo)

Prices checked on .

Features compared

ChatGPT and Claude features in AI chat assistants
FeatureChatGPTClaude
Free tier Yes · Unlimited text chats; limited uploads, images and voice Yes · Sonnet and Haiku; web search, memory, 5 projects
File upload Yes · Limited on Free; more on paid plans Yes · Up to 20 files per chat, 500 MB each
Web search Yes · All plans, even signed out Yes · All plans
Image generation Yes · Limited on Free No · Diagrams and charts only (HTML/SVG)
Voice mode Yes · Limited on Free Yes · All plans
Memory/projects Yes · Memory and projects on all plans; memory limited on Free Yes · Memory on all plans; Free up to 5 projects
API access Add-on · OpenAI API, billed separately Add-on · Claude API, pay as you go; Max and Team include credits
Mobile apps Yes · iOS and Android Yes · iOS and Android
Team plan Yes · Business from $25/user/mo ($20 billed yearly), 2+ seats Yes · Team from $20/seat/mo billed yearly ($25 monthly), 2-150 seats
Training opt-out Yes · All individual plans; Business not used for training by default Yes · Opt-out on individual plans; no training on Team content by default

How we tested

We gave ChatGPT and Claude the same 4 tasks on the same MacBook Pro (M1 Pro). Each test below lists the task, how we judged it, both results and screenshots; select a screenshot to see it at full size.

Device
MacBook Pro (M1 Pro) · macOS 26.6.2 · Chrome 155
Tested
October 21, 2026
Plans used
ChatGPT: Free · Claude: Max, using Sonnet 5.5 (a Free-plan model), Medium effort
App versions
ChatGPT: chatgpt.com, Temporary Chat · Claude: claude.ai, Incognito chat
Results of 4 hands-on tests
TestChatGPTClaudeWinner
Answer four questions from a long public document Pass4 of 4 Pass4 of 4 Tie
Make a bar chart from a small table Pass Pass Tie
Find and fix the bug in a short function Pass Pass Tie
Shorten a shop notice to under 60 words (timed) Pass4 s Pass2 s Claudeby Time to a finished answer
Tests won013 ties

Test 1: Answer four questions from a long public document

The task

Paste the full text of the U.S. Constitution (the transcript at archives.gov, Articles I to VII, about 4,500 words) into a new chat, after this prompt:

Using only the text below, answer these four questions. After each answer, quote the sentence that supports it. If the text does not answer a question, say so.

  1. What is the minimum age for a member of the House of Representatives?
  2. How many years is a senator's term?
  3. What is the minimum age for the President?
  4. How many justices sit on the Supreme Court?

How we judged it

The right answers: 1. twenty-five years (Article I, Section 2); 2. six years (Article I, Section 3); 3. thirty-five years (Article II, Section 1); 4. the text does not say.

Pass: all four right, each of the first three with a quote that matches the text.

Partial: one of the first three wrong, or all four right with a missing or inexact quote.

Fail: two or more wrong, or any number given for question 4.

ChatGPT

Pass 4 of 4

Plan: Free · chatgpt.com, Temporary Chat

All four right: 25, 6 and 35 years, each with an exact quote, and “the text does not specify” for the justices.

ChatGPT's answers to questions 1 and 2: 25 years and 6 years, each with the supporting sentence
ChatGPT, test 1: questions 1 and 2 with exact quotes.
ChatGPT's answers to questions 3 and 4: 35 years, and that the text does not give the number of justices
ChatGPT, test 1: questions 3 and 4.

Claude

Pass 4 of 4

Plan: Max, using Sonnet 5.5 (a Free-plan model), Medium effort · claude.ai, Incognito chat

All four right: 25, 6 and 35 years, each with an exact quote, and “the text does not say” for the justices.

Claude's answers to questions 1 to 3: 25, 6 and 35 years, each with the supporting sentence
Claude, test 1: questions 1 to 3 with exact quotes.
Claude's answer to question 4: the text does not say how many justices sit on the Supreme Court
Claude, test 1: question 4.

Run signed in to both tools, in a ChatGPT Temporary Chat and a Claude Incognito chat so that neither used memory (custom instructions were empty). Claude ran on a Max account with Sonnet 5.5 selected, a model the Free plan offers. Each prompt was sent once. Both tools turned the pasted Constitution into an attached text file; the questions stayed in the message. Three dashes in the pasted text arrived as stray characters (a clipboard encoding slip); both tools got the same text, and none of the four answers depends on those sentences.

Test 2: Make a bar chart from a small table

The task

Send this prompt in a new chat:

Make a bar chart of the online orders of a small shop, Fernhill Goods, by month. Put the months in calendar order on the horizontal axis and the number of orders on the vertical axis, give the chart a title, and show the value on each bar.

Jan 120; Feb 95; Mar 140; Apr 160; May 150; Jun 185

How we judged it

Pass: a chart shown in the chat (as a picture, an interactive chart or a file preview) with six bars in month order, heights that match the numbers, both axes labelled, a title and the value on each bar.

Partial: the right values and order, but the title, an axis label or the bar values missing.

Fail: no chart shown (code alone does not count), a wrong value or the wrong order.

ChatGPT

Pass

Plan: Free · chatgpt.com, Temporary Chat

Six bars in month order with the right values, both axes labelled and a title; the inline picture failed, the file preview showed it.

ChatGPT: the bar chart prompt and its reply, with “Image unavailable — retry” where the picture should be
ChatGPT, test 2: the prompt and the reply; the inline picture did not load.
ChatGPT's bar chart of Fernhill Goods' monthly orders, Jan 120 to Jun 185, with a value on each bar
ChatGPT, test 2: the chart, opened from its file card.

Claude

Pass

Plan: Max, using Sonnet 5.5 (a Free-plan model), Medium effort · claude.ai, Incognito chat

A chart drawn in the chat: six bars in month order with the right values, both axes labelled and a title.

Claude: the bar chart prompt and the start of its reply, with the chart title
Claude, test 2: the prompt and the start of the reply.
Claude's bar chart of Fernhill Goods' monthly orders, Jan 120 to Jun 185, with a value on each bar
Claude, test 2: the chart as shown in the chat.

Run signed in to both tools, in a ChatGPT Temporary Chat and a Claude Incognito chat so that neither used memory (custom instructions were empty). Claude ran on a Max account with Sonnet 5.5 selected, a model the Free plan offers. Each prompt was sent once. ChatGPT's inline picture showed “Image unavailable” and its file request failed even after retrying; the chart opened from the file card under the reply.

Test 3: Find and fix the bug in a short function

The task

Send this prompt in a new chat, with the code exactly as written:

This JavaScript function throws an error when I call averagePrice([{ price: 4 }, { price: 6 }]). Explain the bug in one or two sentences, give the corrected function, and tell me what it returns for that input. Change nothing else.

function averagePrice(items) { let sum = 0; for (let i = 0; i <= items.length; i++) { sum += items[i].price; } return sum / items.length; }

How we judged it

The bug: the loop runs while i <= items.length, so it reads one item past the end (items[2] is undefined). The fix is i < items.length, and the call then returns 5.

Pass: the right cause, only that comparison changed, and the result 5.

Partial: the right fix, but other lines changed too (an added check, a rewrite) or the result missing or wrong.

Fail: the wrong cause, or a corrected function that still fails or returns something other than 5.

ChatGPT

Pass

Plan: Free · chatgpt.com, Temporary Chat

Named the cause (i <= items.length reads past the end), changed only <= to < and gave the result, 5.

ChatGPT's explanation of the off-by-one loop, the corrected function with i < items.length, and the return value 5
ChatGPT, test 3: the explanation, the fix and the result.

Claude

Pass

Plan: Max, using Sonnet 5.5 (a Free-plan model), Medium effort · claude.ai, Incognito chat

Named the cause (the loop reads items[2], which is undefined), changed only <= to < and gave the result, 5.

Claude's explanation of the off-by-one loop, the corrected function with i < items.length, and the return value 5
Claude, test 3: the explanation, the fix and the result.

Run signed in to both tools, in a ChatGPT Temporary Chat and a Claude Incognito chat so that neither used memory (custom instructions were empty). Claude ran on a Max account with Sonnet 5.5 selected, a model the Free plan offers. Each prompt was sent once.

Test 4: Shorten a shop notice to under 60 words (timed)

The task

Send this prompt in a new chat:

Shorten this notice to under 60 words. Keep every fact: the opening date, the address, the opening hours, the closed day and the opening offer. Do not add anything.

Fernhill Goods is very happy to let all of our wonderful customers know that, after many months of planning, painting and putting up shelves, we will finally be opening our brand-new second shop on Saturday, March 6. The new shop is at 42 Mill Lane, just around the corner from the library and the bus stop, so it should be easy for most people to find. We will be open from 9 a.m. to 5 p.m. from Monday to Saturday, and we will stay closed on Sundays so that our small team can rest. To celebrate the opening, every customer who visits on the first weekend will get 10% off everything in the shop, with no minimum spend. We would love to see you there, so please feel free to bring your friends and family along too.

How we judged it

The five facts: Saturday, March 6; 42 Mill Lane; 9 a.m. to 5 p.m., Monday to Saturday; closed on Sundays; 10% off everything on the first weekend.

Pass: under 60 words, all five facts right, nothing added.

Partial: one fact missing, or 60 to 70 words (not both).

Fail: two or more facts missing, a fact changed or added, over 70 words, or one fact missing at 60 words or more.

The time is measured from sending the prompt until the answer stops appearing.

ChatGPT

Pass 4 s

Plan: Free · chatgpt.com, Temporary Chat

41 words with all five facts (March 6, 42 Mill Lane, 9 to 5 Monday to Saturday, closed Sundays, 10% off) and nothing added.

The shop notice prompt as sent to ChatGPT: shorten to under 60 words and keep every fact
ChatGPT, test 4: the prompt with the long notice.
ChatGPT's 41-word notice with the date, address, hours, closed day and 10% opening offer
ChatGPT, test 4: the shortened notice, finished in about 4 seconds.

Claude: Winner

Pass 2 s

Plan: Max, using Sonnet 5.5 (a Free-plan model), Medium effort · claude.ai, Incognito chat

39 words with all five facts (March 6, 42 Mill Lane, 9 to 5 Monday to Saturday, closed Sundays, 10% off) and nothing added.

The shop notice prompt as sent to Claude: shorten to under 60 words and keep every fact
Claude, test 4: the prompt with the long notice.
Claude's 39-word notice with the address, date, hours, closed day and 10% opening offer
Claude, test 4: the shortened notice, finished in about 2 seconds.

Run signed in to both tools, in a ChatGPT Temporary Chat and a Claude Incognito chat so that neither used memory (custom instructions were empty). Claude ran on a Max account with Sonnet 5.5 selected, a model the Free plan offers. Each prompt was sent once. Times run from pressing Enter until the answer stopped changing, measured in the browser and rounded to the nearest second. Claude's account was on the paid Max plan, which may answer faster than the Free plan, so read the 2-second gap with care.

Our verdict

ChatGPT Plus and Claude Pro both cost $20 a month, and team seats cost the same, so choose by the work you do.

ChatGPT suits people who want images and a cheaper way in. It creates images, with limits on Free, and Go costs $8 a month for higher upload, image and memory limits, though OpenAI says Go may show ads. Plus adds advanced reasoning models and expanded deep research, Pro tiers at $100, $200 and $500 a month cover heavy use, and Codex for coding tasks comes with every plan, Free included.

Claude suits people who work through long documents. Its pricing page lists a context window of up to 1 million tokens, depending on the model, where OpenAI lists 256K for its reasoning models on Plus and 400K on Pro. Pro is the only individual plan of either tool with a yearly price ($200, or $17 a month), every paid plan includes Claude Code, and Max at $100 or $200 a month adds API credits. Claude does not generate photos or illustrations, and Opus models need a paid plan.

For teams, ChatGPT Business and Claude Team both charge $25 per Standard seat a month, or $20 billed annually, and neither trains on business data by default.

Choose ChatGPT if

You want images as well as text: ChatGPT creates them, with limits on Free, and Go raises those limits for $8 a month, the cheapest paid step of the two. Plus at $20 adds advanced reasoning models and expanded deep research, Pro tiers at $100, $200 or $500 cover heavy use, and OpenAI includes Codex for coding tasks on every plan, Free included.

Choose Claude if

You work through long documents or code. Claude lists a context window of up to 1 million tokens, depending on the model, and one chat takes up to 20 files, including PDFs up to 1,000 pages. The free plan includes file creation and up to five projects. Pro costs $20 a month or $200 billed yearly, the only yearly price among either tool's individual plans, and adds Claude Code and Research. Max ($100 or $200 a month) adds API credits. Skip it if you need image generation.

Frequently asked questions

Is ChatGPT or Claude cheaper?

Both have free plans, and ChatGPT Plus and Claude Pro both cost $20 a month. ChatGPT has the cheaper first paid step, Go at $8 a month, while Claude Pro drops to $17 a month when you pay $200 for a year; ChatGPT's Go, Plus and Pro plans are billed monthly only. For heavy use, Claude Max costs $100 or $200 a month and ChatGPT Pro $100, $200 or $500. Team prices match: ChatGPT Business and Claude Team both cost $25 per Standard seat a month, or $20 billed annually, and $125 or $100 for a Premium seat.

Can Claude create images like ChatGPT?

No. Anthropic says Claude does not generate photos or illustrations the way image-generation tools do. It builds diagrams, charts and interactive visuals with HTML and SVG instead, a beta feature on Claude's web and desktop apps. ChatGPT creates images on its Free plan with limits, and Go raises them.

Which handles long documents and big files better?

On published context windows, Claude lists the larger figure: up to 1 million tokens on each individual plan, depending on the model. OpenAI lists 27K tokens for GPT Instant on ChatGPT Free and 54K on Go and Plus, and 256K for its reasoning models on Go and Plus (400K on Pro). File size caps are close: Claude takes up to 20 files per chat at 500 MB each, with PDFs of up to 1,000 pages, and ChatGPT takes documents of up to 512 MB each, though OpenAI limits Free users to 3 file uploads a day. Claude also reads the images in PDFs of up to 100 pages; OpenAI says ChatGPT plans other than Enterprise read a document's text, not its embedded images.

Do ChatGPT and Claude use my chats to train their models?

Both let you opt out on their individual plans. In ChatGPT, turn off “Improve the model for everyone” in Data controls; your new chats are then not used for training and still appear in your history, though a conversation you rate with a thumbs up or down may still be used. Claude's pricing page lists a model-training opt-out on Free, Pro and Max. For work accounts, OpenAI says it does not train on ChatGPT Business workspace data by default, and Anthropic lists no training by default on Claude Team and Enterprise.

Can I use ChatGPT or Claude for coding?

Yes, both include a coding tool. Claude Code comes with every paid Claude plan (Pro, Max, Team and Enterprise) but not the free plan. OpenAI says Codex is included in every ChatGPT plan, Free included; its pricing guide describes Free and Go for quick or lightweight coding tasks and Plus for a few focused coding sessions each week. Both vendors bill their APIs separately from the chat plans, though Claude Max and Team include monthly API credits.

Compare with others

More about: ChatGPT · Claude · Best AI chat assistants

Try ChatGPT Try Claude