Claude vs ChatGPT 2026: Which One to Use, Task by Task

The Short Answer

Choose by what you do most often. ChatGPT covers more ground, including images, which Claude does not make. Claude gives you more room for long documents on paid plans, and it does well on the one public writing ranking cited here. For writing and coding, the evidence is close enough that a short test of your own will tell you more than any article.

If you mostly want to...	Start with	Main reason
Make or edit images	ChatGPT	Claude has no native image generation
Talk to an assistant that can see through your camera	ChatGPT	Both have voice, but only ChatGPT lists voice with video
Spend as little as possible	ChatGPT Free or Go	Go costs $8 per month in the US, and Free includes limited deep research. Both plans may show ads
Work through long documents	Claude Pro	Anthropic lists a 1M token chat context on newer models, compared with 256K on ChatGPT Plus with reasoning
Draft long-form or creative writing	Claude Pro, then compare	Opus 5.5 ranks well above GPT-5.6 Sol and GPT-6.1 Sol on Arena's writing board, but this is thin evidence
Code with an AI agent	Test both	Results are close, and they depend on the benchmark. Claude Code needs a paid plan, while Codex has limited desktop access on Free
Hand off multi-step tasks	Test both	Claude Cowork and ChatGPT Work overlap heavily, and both change often
Keep using a custom GPT	ChatGPT, with a migration plan	Custom GPTs are scheduled to retire on December 11, 2026
Live in Afghanistan, Myanmar or Yemen	ChatGPT	These three are on ChatGPT's country list but not on Claude's new list
Build an app rather than chat	See the app section below	A chatbot may not be the right tool

How this comparison was made. This article is based on official product, pricing and help pages, plus public leaderboards, all checked on November 1, 2026. It is not a NxCode test. NxCode did not run these products against each other. An earlier version of this page described a hands-on test, which never took place, so that claim has been removed. There is also no overall winner column, because the sources do not support one. NxCode is our own product and it appears in one section below, marked as such.

Where official pages disagree with each other, this article says so instead of picking a side. Plans and limits change often, so confirm anything that matters on the provider's page before you pay.

What Each Product Is

Claude

Claude is made by Anthropic and runs on the web, on desktop, and in iOS and Android apps. The current model line has four names: Haiku 5.5 (the small one), Sonnet 5.5, Opus 5.5 and Fable 5.1. Which you can use depends on your plan:

Free: Sonnet and Haiku, with no Opus or Fable. The page names only the families, not the versions.
Pro: Opus, Sonnet and Haiku. Fable is paid for with usage credits.
Max and Ultra: all of those, with Fable taking up to 50% of your weekly allowance.

Anthropic does not publish which model the chat app selects by default. Mythos 5.1 is limited to approved partners, so regular users cannot pick it.

Around the chat sit Research, Artifacts, Docs, Slides, Design, Skills, connectors, Cowork for longer tasks, and Claude Code for programming.

ChatGPT

ChatGPT is made by OpenAI and has three separate areas: Chat, Work and Codex. They use different models, which is the main source of confusion.

Chat: on Plus, Pro, Business and Enterprise the default is GPT-6, still rolling out, powered by GPT-6 Sol. The Pro option uses GPT-6 Astra and is not on Plus. GPT-5.6 Sol can still be selected from the menu.
Work and Codex: GPT-6.1 Sol, GPT-6.1 Terra, GPT-6 Sol and GPT-6 Luna. Plus gets limited GPT-6 Astra here, then pays with credits, but does not get Astra in Chat.
Free and Go: OpenAI's pages conflict. The release notes say these plans use GPT-6 Luna, while the model help article says the default is GPT-5.6 Luna. Check what your own account shows.

Several older names are out of date. The ChatGPT agent has been taken offline and OpenAI now points to ChatGPT Work. The Sora web and mobile apps stopped on April 26, 2026. GPT-5.5 left ChatGPT on October 14, 2026.

Terms to Know

A few words decide most of this comparison. Here is what each one means.

Token: a small piece of text, shorter than most words. Prices and memory limits are counted in tokens.
Context window: how much text the assistant can hold in view during one chat, including your files and its replies. Think of it as the size of its desk.
Effort or thinking level: how hard the model works on a reply. More effort usually means better answers on hard questions, slower replies and faster use of your allowance.
Agent: an AI that takes a series of steps toward a goal, such as editing files or using a website, instead of giving one reply.
Connector or plugin: a link between the assistant and another service, such as your email or calendar.
Opt out: a setting you change yourself so your chats are not used to train future models.

Claude vs ChatGPT: Comparison Table

This table describes what each company lists. It is not a score.

Feature	Claude	ChatGPT
Cheapest paid plan	Pro, $20/mo or $17/mo billed yearly	Go, $8/mo in the US
Models on Free	Sonnet and Haiku	Official pages conflict: GPT-6 Luna or GPT-5.6 Luna
Image generation	Not supported	Free limited, Go and above
Video generation	Not supported	No such row on the pricing page; Sora stopped
Voice chat	All plans (beta)	All plans, limits vary
Voice with video	Not listed	Go and above
Web search	All plans	All plans
Deep research	Research, Pro and above	Free and Go limited, Plus, Pro highest
Memory	All plans	Free limited, Go, Plus and Pro expanded
Search past chats	Paid plans	Memory with past chats: Free limited, Go, Plus and Pro expanded
Projects	Free up to 5, Pro and above	Feature table lists all plans; plan card presents it as a Plus addition
Project sharing	Team and Enterprise only	All plans
Chat context window	Up to 1M tokens on newer models	Plus 54K Instant, 256K reasoning; Pro 128K and 400K
Upload size per file	500MB in chat, 30MB in projects	512MB; spreadsheets about 50MB, images 20MB
Coding agent	Claude Code, Pro and above	Codex: limited desktop app on Free and Go, more surfaces from Plus
Long-running tasks	Cowork, Pro and above	ChatGPT Work: limited desktop on Free and Go, all surfaces on Plus and Pro
Scheduled tasks	Pro and above	Go and above; Free is disputed between official pages
Control your computer	Pro, Max, Ultra (beta)	Computer Use in desktop Work and Codex; plan list not stated
Reusable instructions	Skills, all plans	Skills, all plans; custom GPTs retiring December 11, 2026
Mobile apps	iOS and Android, all plans	iOS and Android, all plans

Writing: What the Public Rankings Show

There is no neutral measurement of "better writing." Claims that one product simply writes better are usually opinion. The public reading available is the Arena creative writing board, where scores come from votes by real people. The data date is October 8, 2026, with about 1.39 million votes across 412 models, read from an October 11 archived copy. The table shows selected rows.

Rank	Model setting	Score	Rank range
1	Gemini 4 Argon (High)	1519 ±20, marked Preliminary	1–5
2	Claude Opus 5.5 (High)	1517 ±17	1–5
3	Claude Fable 5 (High)	1502 ±8	1–8
20	GPT-5.6 Sol (xHigh)	1468 ±8	10–37
24	Claude Sonnet 5.5 (xHigh)	1463 ±19	6–63
26	GPT-6.1 Sol (Max)	1462 ±20	7–65
89	GPT-6 Luna (Max)	1411 ±9	73–123

Four things to take from this table:

Gemini 4 Argon sits at the top, and it belongs to neither product. Its score is marked Preliminary, and its rank range overlaps Opus 5.5's.
Opus 5.5 and Fable 5 sit well above GPT-5.6 Sol and GPT-6.1 Sol on this board. The rank ranges for Opus 5.5 and GPT-5.6 Sol do not overlap.
Claude's Sonnet 5.5 shows no such gap. Sonnet 5.5 and GPT-6.1 Sol land next to each other, with wide, overlapping ranges.
These are model settings, not app plans. Claude Free has no Opus, so this reading does not describe what a free Claude user gets. A setting like "(High)" may also differ from what your plan gives you.

The overall text board (data date October 20, about 8.9 million votes) follows a similar order: Gemini 4 Argon first at 1524 ±9 (Preliminary), Opus 5.5 third, GPT-5.6 Sol 21st and GPT-6.1 Sol 28th. Both boards are snapshots, and neither is today's live ranking.

Example: you have a rough 300-word product announcement. Paste the same draft into both, ask for the same length and audience, and count how many sentences you would still change. That number says more about your own use than a leaderboard does.

Coding: Claude Code vs Codex

Both companies include a coding agent in their paid subscriptions.

Claude Code runs in the terminal, in VS Code and JetBrains, in a desktop app, and on the web and mobile. It defaults to Opus 5.5 on every paid plan. It is not included in Free and it is not open source.

Codex runs in a desktop app, a command line tool, IDE extensions, the web and iOS, with GitHub code review and Slack integrations. Free and Go get the desktop app with GPT-6 Luna, subject to rollout. Plus and above get the other surfaces. Its command line tool is open source under Apache-2.0, though the IDE extension and Codex cloud are not.

So for a beginner with no budget, Codex is the only one of the two you can try for free. OpenAI also publishes rough figures for Plus, such as 15–160 local messages per 5 hours on GPT-6.1 Sol, and it stresses these are estimates. Anthropic does not publish a matching number for Pro.

Public coding benchmarks are less settled than they were. OpenAI stopped reporting SWE-bench Verified in February 2026, after an audit found flawed test cases in at least 59.4% of the problems it reviewed. In July it withdrew its recommendation of SWE-Bench Pro too, estimating that about 30% of tasks had problems. This page no longer quotes the SWE-bench scores it used to carry. Newer readings look like this:

Evaluation	Run by	Claude	OpenAI
FrontierCode 2.0	Cognition	Opus 5.5: 45.5%	GPT-6 Astra: 42.8%
SWE-Bench Pro V2, hard view	Scale	Opus 5 in Claude Code: 98.00	GPT-6 Astra in Codex: 90.20
SWE-Bench Pro V2, full view	Scale	Opus 5 in Claude Code: 99.40	GPT-6 Astra in Codex: 96.90
DeepSWE v1.1	Datacurve	claude-opus-5: about 74%	gpt-6-astra: about 74%

Claude is ahead on FrontierCode and both Scale views and level on DeepSWE, but the caveats are real. The boards use different methods and cannot be compared with each other. Opus 5.5 is not on Scale's board, so those rows show an older model. FrontierCode is run by a company that sells its own coding product. Gemini 4 Argon sits between the two companies on both Scale views. A fair reading is "close, with Claude ahead on some boards," not a clear win.

For a deeper look at coding tools, including options beyond these two, see our ranking of AI for coding.

Research and Web Search

Both products search the web on every plan, so old claims that Claude has only limited web access no longer hold. In Claude's newer experience there is no search toggle. It searches when needed, and searches count against your limits. On Team and Enterprise, an Owner has to enable search first.

The difference is in longer research reports.

Claude Research needs Pro or above. It always runs on Opus 5.5, falling back to Opus 4.8, no matter which model you selected. It uses your allowance faster than normal chat, and when started on a phone it searches the web only, without your connectors.
ChatGPT deep research is on Free and Go in limited form, on Plus, and at the highest level on Pro. OpenAI does not publish the counts and points you to the in-product counter.

Example: ask both to compare three health insurance options in your city, then open every link. A confident summary with a source that does not say what the summary claims is a failure, whichever product wrote it.

Images and Video

This is the clearest difference. In Anthropic's words, "Claude doesn't generate photos or illustrations the way image-generation tools do." Claude can analyze images you upload, and can draw charts and interactive visuals with HTML and SVG in the web and desktop apps (beta).

ChatGPT ships ChatGPT Images 2.5, released September 8, 2026, with templates, generation from sketches, and comment-style editing on mobile. It is limited on Free and available on Go and above, and Image generation with Thinking needs Plus or Pro. OpenAI does not publish image counts per plan.

Neither product is a realistic video generator on the pages reviewed. Sora's apps have stopped, and ChatGPT's pricing table has no video generation row. "Voice with video" means ChatGPT can see through your camera, not that it makes videos. Claude Motion exports short animations as MP4 on Team and Enterprise (beta), but Anthropic says it "doesn't use a video generation model."

Voice

Older comparisons say Claude has no voice mode. That is wrong now. Claude voice is in beta on all plans, including Free, on iOS, Android, desktop and web. Anthropic says it is built to work best on a phone. It offers hands-free and push-to-talk, can use your connected tools, and counts against normal usage. Fable models do not support voice.

ChatGPT Voice has three modes. Live listens and talks at the same time and can use search, memory and plugins. Advanced handles mobile video and screen sharing. Standard is the third. OpenAI publishes rolling 24-hour limits for Chat:

Plan	ChatGPT voice limit
Free	Limited, no number published
Plus	3 hours
Pro $100	15 hours
Pro $200	Unlimited, subject to abuse guardrails
Business Standard / Premium	3 hours / 15 hours, then credits

Go is left out because OpenAI's pages disagree: the Go article says the same limits as Free, the Voice article says 3 hours, and the pricing page says "Expanded." Pro 500 is not listed in that table.

For most people the practical gap is the camera. If you want to point your phone at something and ask about it, ChatGPT lists that on Go and above, and Anthropic's pricing page lists nothing equivalent.

Files, Long Documents and Projects

Context window. Anthropic lists 1M tokens in chat on paid plans for Fable 5.1, Opus 5.5, Opus 5, Sonnet 5.5, Sonnet 5 and Haiku 5.5, and 500K on some older models. Its pricing page says "Up to 1M varies by model" for Free too, though one Team help article still says 200K. ChatGPT's pricing page gives smaller figures for Chat:

ChatGPT Chat	Free	Go and Plus	Pro
Instant window	27K	54K	128K
Instant input, approx.	12 pages	40 pages	250 pages
Reasoning window	Varies	256K	400K
Reasoning input, approx.	Varies	320 pages	680 pages

OpenAI notes that usable space is smaller than the total and shifts with features and memory. Work and Codex differ: GPT-6.1 Sol has a 500,000 token window there. The two companies describe these limits differently, and Anthropic says its usage limits vary with chat length and file size, so treat these figures as a guide rather than a like-for-like test.

Uploads. Claude takes up to 20 files per chat at 500MB each. PDFs can run to 1,000 pages, but images are analyzed only up to 100 pages. Above that, only text is read. Its documentation does not list audio or video formats. ChatGPT takes 512MB per file, up to 2 million tokens per document, and 80 files every 3 hours, with 3 per day on Free (a September snapshot; limits can drop at peak times). Audio uploads came to paid plans on October 6.

Projects. Claude Free gets up to 5 projects. Project files are 30MB each, with no set count as long as they fit the context, and sharing is only on Team and Enterprise. ChatGPT projects allow 5 files on Free, 25 on Go and Plus, and 40 on Pro and business plans, and you can share with 5 people on Free, 10 on Go and Plus, and 100 on Pro.

Documents and pages. Claude has Artifacts, Docs, Slides and Design on all plans, and Artifacts can be published and shared. ChatGPT has Sites on Plus and Pro, and Space with Pages and Slides on Pro and above. If you relied on ChatGPT Canvas, check your account first: it was removed from GPT-5.5 in May, and its status under GPT-6 could not be confirmed.

Memory

Claude memory is on by default for Free, Pro, Max and Ultra. On Team and Enterprise, an Owner controls it and members turn it on themselves. You can edit what it knows under Settings > Memory > Topics, and sensitive subjects like health and beliefs are not saved by default. Searching past chats is limited to paid plans. Incognito chats are available on every plan.

ChatGPT memory is limited on Free, available on Go, and expanded on Plus and Pro, and past-chat memory follows the same pattern. OpenAI says exact abilities vary by plan, region and platform, and gives no numbers. Temporary chats skip personalization by default. A project can use project-only memory, but that mode blocks ChatGPT Work.

Example: tell each one your dietary needs, start a new chat a day later, and ask for a meal plan. Then look at what was stored and delete anything you would not want kept.

Automation and Agents

Both companies now offer an assistant that keeps working on a task after you stop typing.

Capability	Claude	ChatGPT
Long tasks	Cowork, Pro and above; new Pro, Max and Ultra tasks run in the cloud	ChatGPT Work: limited desktop on Free and Go; web and mobile for eligible paid plans
Using a browser	Claude in Chrome, paid plans	Built-in browser on all plans; Work's cloud browser for logged-in sites is Plus and Pro
Using your computer	Pro, Max, Ultra beta on macOS and Windows	Computer Use in desktop apps on macOS and Windows, supported regions
Scheduled tasks	Pro and above	Go and above; for Free, the pricing table says no but August release notes say 3 active tasks
Always-on agent	Duet, research preview, US waitlist, Pro, Max and Ultra	Dots, on Pro, Business Premium and Enterprise

Custom GPTs are part of this picture. OpenAI plans to retire them on December 11, 2026 on all plans, with February 11, 2027 for FedRAMP and approved Enterprise workspaces. Moving one to a plugin turns its instructions into a skill, but the chosen model is not kept and custom actions do not move automatically. If you depend on one, test the migration before that date.

Both products use Skills. Claude's need code execution turned on. Plugins are on Claude's paid plans and on all ChatGPT plans, where creating and sharing them starts at Go. Claude Free allows one custom connector. Whether that cap also covers connectors from Anthropic's directory is not clear from its pages.

An agent that can act for you carries more risk than a chat. Permission matters: give it the smallest access that works, and watch the first few runs before leaving it alone with your accounts.

If You Want to Build an App, Not Just Chat

Sometimes the real goal is a working tool, such as a booking page or an internal dashboard, and a chat window is only a means to that. Claude Artifacts can be published and shared, and ChatGPT offers Sites on Plus and Pro. Claude Code and Codex are aimed at people who are comfortable with code and tests.

NxCode, our own product, is an AI app builder for people who want to go from a written brief to a working app without managing a coding agent. It is a different kind of tool from Claude or ChatGPT, not a replacement for a general assistant. Since NxCode is ours, judge it the same way as anything else here: by whether the finished app does what you asked. For more choices, see our comparison of AI app builders.

Pricing: Plans and API

Prices are in US dollars before tax. Sources: Claude pricing, ChatGPT and Codex pricing, ChatGPT Pro tiers, ChatGPT Business.

Claude plan	Price	Usage as described
Free	$0	Limited, numbers not published
Pro	$20/mo, or $17/mo billed yearly ($200)	More than Free, numbers not published
Max	$100/mo	5x Pro
Ultra	$200/mo	10x Pro weekly; no 5-hour limit on verified accounts
Team Standard	$25/seat, or $20 yearly	2–150 people, more than Pro
Team Premium	$125/seat, or $100 yearly	5x Team Standard
Enterprise	$20/seat yearly, plus usage at API rates	20 seats minimum

ChatGPT plan	Price	Usage as described
Free	$0	Unlimited text chats within abuse limits; tools limited; may show ads
Go	$8/mo in the US, local pricing elsewhere	Unlimited text chats, more tool use than Free; may show ads
Plus	$20/mo	"Limits apply"; no ads
Pro	$100, $200 or $500/mo	Higher limits at each step; Ultrafast only at $500; no ads
Business Standard	$25/seat, or $20 yearly	At least 2 seats
Business Premium	$125/seat, or $100 yearly	More Work and Codex use, no 5-hour limit
Enterprise and Edu	Contact sales	Not published

Some details change the real cost:

Billing: ChatGPT Go, Plus and Pro have no annual option, and Claude Max and Ultra are monthly only. Claude prices may differ on mobile.
Multipliers: Anthropic never says how many messages "1x" equals, and it changes with model, chat length and file size. OpenAI has removed the older Pro multipliers from its help page.
Running out: paid Claude plans can buy usage credits to keep going.
Included API credit: Max's $100 and Ultra's $200 monthly credits are for building on the Claude Platform. They do not cover chat or interactive Claude Code.
Ads: OpenAI says it does not share chats with advertisers or sell data to them. Ads are off for minors, and you can turn off ad personalization in settings.

If you are a developer paying by usage, the API is a separate bill. Prices per million tokens, input then output:

Tier	Claude	OpenAI
Largest	Fable 5.1: $10 / $50	GPT-6 Astra: $10 / $50
Main	Opus 5.5: $4 / $20; Sonnet 5.5: $2 / $10	GPT-6.1 Sol: $2 / $10; GPT-6.1 Terra: $1 / $5
Small	Haiku 5.5: $0.10 / $0.50 up to 100K prompt tokens, then $0.50 / $2.50	GPT-6 Luna: $0.10 / $0.50

OpenAI's figures are for inputs up to 272K tokens, from its API pricing page. These tiers are a rough grouping, not official equivalents.

Privacy and Data Training

On personal plans, whether your chats train future models depends on a setting you control. Check it yourself, because a pricing page label does not tell you what your account is set to.

Question	Claude	ChatGPT
Personal plans	Pricing page says "Opt-out"; help center says used if "You choose to allow us"	Pricing page says "Opt-out available"; help center says "we may use your content to train our models"
Where to change it	Settings > Privacy > Help Improve our AI models	Settings > Data controls > Improve the model for everyone, or the privacy portal
Chats not used for training	Incognito chats	Temporary chats, kept up to 30 days
Thumbs up or down	Stored 5 years, may be used for training	The whole conversation may be used for training
Storage	Up to 5 years if allowed (de-identified), 30 days if not; deleted chats removed within 30 days	Deleted chats removed within 30 days, except for legal or security reasons
Connected apps	Raw connector content not used unless copied into chat	App data may be used if the setting is on
Business plans	Team, Enterprise and API not used by default	Business, Enterprise, Edu and API not used by default

Three points are easy to miss:

Anthropic's 2025 terms update made this a choice at sign-up, but its pages do not say whether the option is preselected. Open the setting and check.
Opting out with Anthropic applies to future training. Training already under way is not undone. Chats flagged by safety systems can still feed internal safety models.
OpenAI's opt-out covers ChatGPT and Codex tasks, but Codex has a separate Include environments setting that it does not change.

A sensible rule for any personal account: do not paste client data, health records or passwords into a chat you would not want stored. If your company handles confidential data, use a business plan and read its security terms.

Regional Availability

Check this before paying, because the two lists differ.

Claude: the supported regions policy lists 185 countries and regions, effective November 12, 2026, so that version is not yet in force today, and the text that applies until then could not be confirmed. It also applies to organizations majority-owned from unsupported regions, even when users sit in a supported country. Taiwan is on the list. In Ukraine, Crimea, Donetsk, Kherson, Luhansk and Zaporizhzhia are excluded. You must be 18.
ChatGPT: the supported countries page lists 208 countries and regions, and it counts some territories separately, so the two totals are not directly comparable. Ukraine is listed "with certain exceptions." OpenAI warns that use outside the list may lead to a ban.
Neither: mainland China, Hong Kong, Macau, Russia, Belarus, Iran, Cuba, North Korea, Syria and Venezuela.
ChatGPT only: Afghanistan, Myanmar and Yemen, based on Claude's November 12 list.

Some features have narrower limits. Claude Duet and Claude health data are US only. ChatGPT Finances is US only, and Dots excluded the EEA, Switzerland and the UK at launch. No later update on Dots could be found.

How to Test Both Yourself in 30 Minutes

A short test with your own work beats any comparison, including this one. Use this checklist, and use the same plan level on both sides if you can.

Set up (5 minutes). Open both in separate tabs. Write down your plan, the model shown, and the date. Decide on your training setting before you upload anything, and use non-confidential material.
Document test (10 minutes). Upload the same 10-page report. Ask for the three most important decisions, with page numbers. Check each page number against the original.
Writing test (10 minutes). Give both the same brief, draft and word limit. Ask for one round of changes.
Your must-have feature (3 minutes). Try the one thing you cannot do without, such as voice, image editing or a connector, on your plan and in your country.
Score (2 minutes). Fill in the card below and pick the one that needed less fixing.

Example Scorecard

This is a blank template, not a result.

Check	Claude	ChatGPT
Factual errors found
Page numbers or links that were wrong
Sentences you still had to rewrite
Follow-up messages needed
Hit a usage limit?
Must-have feature worked?

Only compare results from a single sitting. If you change the model, the plan or the prompt partway through, you can no longer say which change made the difference.

Conclusion

The hard differences are few and clear. ChatGPT makes images, sees through your camera in voice, has an $8 tier, and gives free users limited image, deep research and coding access. Claude has a larger chat context on paid plans. Claude Free has no Research and no Claude Code. Custom GPTs are ending, and Sora is gone.

The softer questions are less clear. On the public boards cited here, top Claude models do well for writing and coding, Gemini 4 Argon sits above both on Arena, and none of those readings tell you how the app behaves on your plan.

If you want one to start with: pick ChatGPT if you value range and images, and Claude if your days are spent in long documents and drafts. Then run the 30-minute test before you commit money. If what you really want is an app, look at a builder such as NxCode (ours) or another listed in our app builder comparison.

FAQ

Is Claude better than ChatGPT in 2026? Neither is better at everything. ChatGPT makes images and Claude does not. Claude lists a larger chat context window on paid plans. For writing and coding the public evidence is narrower, so test both on your own work. This answer comes from official pages and public rankings, not from a NxCode test.

Which is better for writing, Claude or ChatGPT? The only public reading used here is the Arena creative writing board (data date October 8, 2026). Gemini 4 Argon was first with a Preliminary label, Claude Opus 5.5 was second, GPT-5.6 Sol was 20th and GPT-6.1 Sol was 26th. That is a vote-based ranking of model settings, not proof of which app will write better for you.

Which is cheaper, Claude or ChatGPT? Both have a free plan and a $20 per month plan. ChatGPT also has Go at $8 per month in the US, which may show ads. Claude Pro drops to $17 per month with annual billing, while ChatGPT Go, Plus and Pro have no annual option.

Can Claude generate images like ChatGPT? No. Anthropic says Claude does not generate photos or illustrations the way image tools do. Claude can analyze images you upload and can build charts and interactive visuals with code. ChatGPT generates and edits images, with limited use on Free.

Does Claude have voice mode? Yes. Claude voice is in beta on all plans, including Free, on iOS, Android, desktop and web. ChatGPT also has voice on all plans, and adds voice with video on Go and above.

Do Claude and ChatGPT train on my chats? They can on personal plans, depending on a setting you control. Anthropic labels Free, Pro, Max and Ultra as Opt-out on its pricing page, and OpenAI labels Free, Go, Plus and Pro as Opt-out available. Business plans from both companies are not used for training by default.

What is happening to custom GPTs? OpenAI plans to retire custom GPTs on December 11, 2026 across all plans, with a later date of February 11, 2027 for FedRAMP and approved Enterprise workspaces. Moving one to a plugin turns its instructions into a skill, but its selected model is not kept and custom actions do not transfer automatically.

Should I pay for both Claude and ChatGPT? Usually not at first. Run the 30-minute test in this article on the free plans, then pay for the one that finished your real task with fewer corrections. Add a second subscription only if you keep needing something the first one lacks, such as image generation.

Sources

All sources were checked on November 1, 2026. Pages on chatgpt.com, help.openai.com, openai.com and arena.ai blocked direct access that day, so those were read through Wayback Machine archive copies. The date beside each one is the snapshot date. Anthropic pages, learn.chatgpt.com and developers.openai.com were read live.

Anthropic

OpenAI, read live

OpenAI, read from archive copies

ChatGPT pricing (October 29)
GPT-6 and other models in ChatGPT (October 29)
ChatGPT release notes and model release notes (October 30)
ChatGPT Work and Codex (November 1)
ChatGPT Voice (October 30)
ChatGPT Go and Pro tiers (October 30)
Ads in ChatGPT (October 22)
Custom GPT retirement (October 30)
Sora discontinuation (October 23)
How your data is used (October 27)
File uploads FAQ (September 18)
Projects in ChatGPT (October 23)
Supported countries (October 21)
Why SWE-bench Verified was dropped (October 15)
Separating signal from noise in coding evaluations (October 20)

Public rankings

Arena text board (archive copy, October 21; data date October 20)
Arena creative writing board (archive copy, October 11; data date October 8)
Scale SWE-Bench Pro V2
FrontierCode 2.0
DeepSWE