ChatGPT vs Claude vs Gemini: Which to Pick for Each Task

As of November 10, 2026, Claude is the app to try first for coding, research, and building presentations. Gemini wins a different test: blind voters prefer it for everyday chat.

ChatGPT is level with Claude on reading long documents and pulling data out of them. Microsoft Copilot can't be scored at all, because Microsoft doesn't name the model it runs.

Two things matter more than that order. The best models sit behind paid plans, and the benchmark leader isn't always the model people prefer.

For any two of the three, the pair pages go deeper: ChatGPT vs Claude, ChatGPT vs Gemini, and Claude vs Gemini. Microsoft's app has its own page, ChatGPT vs Copilot, and a section further down compares all four on plans, features, and what they do with your chats.

Benchmarks pick a winner per job, not one overall winner

We grouped the jobs on our best AI for each job pages by the evidence each one rests on. Then we took the highest-scoring model in each company's current lineup: GPT models for ChatGPT, Claude models for Claude, and Gemini models for Gemini. It's the same comparison the LLM selector shows, and it updates whenever the rankings do.

On the November 11, 2026 data, Claude’s top model scores highest in 7 of the 12 job groups below, Gemini’s leads on finance and accounting work, and math, image understanding, long documents, and data extraction are tied.

Table 1 Download CSV

Job group	ChatGPT	Claude	Gemini
Overall	GPT-6 Astra, 85.0	Claude Fable 5.2, 91.4	Gemini 4 Carbon, 87.9
Knowledge	GPT-6.1 Sol, 84.1	Claude Fable 5.2, 97.4	Gemini 3.8 Flash, 69.2
AA-Omniscience Index	GPT-6 Astra, 42.4	Claude Fable 5.2, 66.8	Gemini 4 Carbon, 53.5
Reasoning	GPT-6 Astra, 80.4	Claude Fable 5.2, 89.9	Gemini 4 Carbon, 83.4
FrontierMath v2 (Tiers 1-3)	GPT-6 Astra, 93.7%	Claude Fable 5.2, 95.1%	Gemini 4 Carbon, 90.5%
AA-MMMU-Pro	GPT-6 Astra, 86.9%	Claude Opus 5.5, 87.7%	Gemini 3.8 Flash, 85.6%
AA-LCR	GPT-6 Luna, 83.3%	Claude Opus 5.5, 84.7%	Gemini 3.1 Pro, 82.0%
Coding	GPT-6 Astra, 73.9	Claude Fable 5.2, 86.3	Gemini 4 Carbon, 80.5
AA-AnalystAgent	GPT-6 Astra, 51.2	Claude Sonnet 5.5, 57.5	Gemini 3.8 Flash, 46.3
Vals Finance Index	GPT-6.1 Sol, 55.8%	Claude Opus 5.5, 65.3%	Gemini 4 Carbon, 67.0%
ExtractBench	GPT-6.1 Sol, 95.3%	Claude Sonnet 5.5, 94.7%	Gemini 3.8 Flash, 80.7%
Agentic	GPT-6.1 Terra, 76.0	Claude Fable 5.2, 90.1	Gemini 3.8 Flash, 64.7

Overall is the AI assistant job. Knowledge is the studying job. AA-Omniscience Index is the research job. Reasoning covers reasoning and RAG. FrontierMath v2 (Tiers 1-3) is the math job. AA-MMMU-Pro covers image understanding and chart analysis. AA-LCR is the PDFs job. Coding covers Excel, coding, debugging, and website building. AA-AnalystAgent is the data analysis job. Vals Finance Index is the finance and accounting job. ExtractBench is the data extraction job. Agentic covers presentations, agents, and customer service.

Bold marks the highest score in each row. Scores sit on our 0 to 100 benchmark scale, and a gap of a point or two won't tell you which app writes your report better. Act on the rows where the columns are far apart, and treat close rows as ties until your own test breaks them.

One clarification, though: each cell is the best model that company sells, not the model your plan opens by default. Free plans leave out the top models, so read this table together with the plan table further down.

Writing has no row. Our pages for writing emails, marketing copy, and creative writing rank models on instruction following, and we leave a job out of this comparison when fewer than two of the three lineups have a measured model for it. Writing is one of those jobs, so the next section uses blind votes instead.

Blind votes measure something benchmarks don't

Arena runs blind tests. Someone types a prompt, two anonymous models answer, and the person picks the reply they prefer. Enough of those votes turn into a rating for each model. A benchmark grades an answer against a key, while Arena records which answer a person liked.

Table 2 Download CSV

Arena category	ChatGPT	Claude	Gemini
Text Overall	GPT-6.1 Sol, 1,482 ±6	Claude Opus 5.5, 1,506 ±6	Gemini 4 Carbon, 1,526 ±10
Creative Writing	GPT-6 Astra, 1,450 ±11	Claude Opus 5.5, 1,518 ±11	Gemini 4 Argon, 1,525 ±10

Each rating comes with a ± margin that Arena publishes. Bold marks the top rating and any rating whose margin overlaps it, so two bold entries in a row are a statistical tie.

Gemini 4 Carbon (1,526 ±10) leads Arena's overall text votes, ahead of Claude Opus 5.5 (1,506 ±6) and ChatGPT's GPT-6.1 Sol (1,482 ±6). Arena scores as of November 9, 2026.

On these numbers the two kinds of evidence point different ways. Graded tests put Claude's best model ahead for most jobs, and the people voting preferred Gemini's.

Neither result is the real one. Arena's prompts are whatever its visitors choose to type, and a vote rewards the reply that person liked better. Benchmarks are narrower and stricter, and they reward the correct answer. If your day is quick questions, trip planning, and rewording messages, the votes are closer to your use. If you hand the app a spreadsheet or a codebase, the graded results are.

Writing: the top-rated models both need a paid plan

For writing, Arena's creative-writing votes are the best evidence we have that covers all three apps.

Gemini 4 Argon (1,525 ±10) and Claude Opus 5.5 (1,518 ±11) are tied at the top of Arena's Creative Writing votes, because their margins overlap. ChatGPT's GPT-6 Astra (1,450 ±11) is further back. Arena scores as of November 9, 2026.

The catch for most readers is that neither model at the top comes free. Claude Free lists Sonnet and Haiku but not Opus, according to the plan registry we keep from Anthropic's pricing page. Google's plan page says Gemini 4 Argon is available to Google AI Pro and Ultra subscribers. On the free tiers, you're comparing Claude Sonnet with Google's lighter Gemini models and with ChatGPT's free plan, and on the October 12 Arena read, Claude Sonnet 5.5 rated about level with ChatGPT's best model for creative writing.

Votes measure preference, not whether a draft follows your brief or keeps your facts straight. For that, test on your own material: give each app the same three briefs and count the edits before each draft is usable. Our writing guide walks through that test.

Gemini has one advantage that no rating captures. Google AI Pro puts Gemini inside Gmail and Docs, according to Google's plan help page. If most of your writing is email replies and documents you already keep in Google's apps, that may matter more than a tie in the ratings.

Coding and research: check the limits, not just the scores

The coding row in the first table covers the jobs most people mean by coding: spreadsheet formulas, data analysis, debugging, and building a website. On the November 10 data, Claude's lineup leads it. The full ranking across every lab is on the coding leaderboard, and the coding job page has the shortlist. If you're choosing between Claude Code and Codex specifically, our ChatGPT or Claude for coding comparison sets independent agent results against what each bug fix costs.

For a paid user, the bigger question is how much use the plan buys. Claude Pro includes Claude Code and meters usage in five-hour sessions with a weekly cap, as our Claude usage limits page explains. OpenAI publishes estimated Codex message ranges per five-hour window for Plus, and the range shifts with the model you pick (Codex usage limits). If you code every day, read those limits before the scores. A plan you exhaust by Wednesday costs you more than a few points of benchmark score.

Studying falls under the knowledge row, and research has its own row, AA-Omniscience Index. Claude's top model leads both on the November 10 data. Both rows grade answers to knowledge questions, so they don't establish whether an app finds good sources or cites them accurately. Check the citations on a topic you know well before you trust any app with one you don't.

The reasoning row covers multi-step problems and answering questions from documents you supply. Claude Fable 5.2 leads it on the November 10 data, but Claude Pro doesn't include Fable in its allowance. Claude Opus 5.5, which Pro does include, scores about level with ChatGPT's best model on that row. If your work is long chains of logic, such as planning a budget with many constraints or checking a contract against a policy, treat the two $20 plans as tied and let your own test decide. Our research job page and studying job page keep the full shortlists.

Everyday use comes down to the free plan and your tools

For quick questions, recipes, travel plans, and rewriting a message, all three apps do the job, and the blind votes above favor Gemini. What separates them for daily use is mostly outside the model.

Table 3 Download CSV

If you mostly...	Start with	Why
Live in Gmail and Google Docs	Gemini	Google AI Pro puts Gemini inside both
Live in Word, Excel, and Outlook	Copilot	Microsoft 365 Personal builds it into the desktop apps
Want the cheapest paid upgrade	Gemini	Google AI Plus is the lowest-priced paid plan we list, though it doesn't include Gemini 4 Argon
Write long documents or code	Claude	Its models lead our coding and knowledge rows
Read long PDFs or pull data out of documents	ChatGPT or Claude	Their models are level on our long-document and data extraction rows

We don't have a pick for voice conversations or image generation. Our voice and image job pages say so and point to the voice benchmarks and image and video pricing instead. A claim that one app wins on voice needs a test of the apps' voice modes, and the model rankings above don't supply one.

Memory, saved projects, and how each app handles your history differ too, and none of those shows up in a benchmark. The features table further down lists what each plan page says about them. If you've built months of context in one app, that's a real switching cost.

What each plan gets you

Prices and model lists below come from our subscription registry, which copies each company's own pricing page, read October 16, 2026 for ChatGPT and Claude, and October 21, 2026 for Gemini.

Table 4 Download CSV

App	Plan	Per month	Models the plan lists
ChatGPT	Free	Free	GPT-6 Luna in Codex; GPT-5.6 Terra with limited access in Work and Codex on desktop.
ChatGPT	Go	$8	GPT-6 Luna in Codex; GPT-5.6 Terra with limited access in Work and Codex on desktop.
ChatGPT	Plus	$20	GPT-6 Astra, GPT-6.1 Sol, GPT-6 Sol and GPT-6 Luna in Codex; GPT-6.1 Sol is available in Work and Codex, not Chat.
ChatGPT	Pro 100	$100	GPT-6 Astra, GPT-6.1 Sol, GPT-6 Sol and GPT-6 Luna with expanded access, plus GPT-5.6 Sol Pro.
ChatGPT	Pro 200	$200	GPT-6 Astra, GPT-6.1 Sol, GPT-6 Sol and GPT-6 Luna with expanded access, plus GPT-5.6 Sol Pro.
ChatGPT	Pro 500	$500	GPT-6 Astra, GPT-6.1 Sol, GPT-6 Sol and GPT-6 Luna with expanded access, plus GPT-5.6 Sol Pro; Ultrafast for GPT-6 Astra and GPT-6.1 Sol.
Claude	Free	Free	Sonnet and Haiku; the plan comparison lists no Opus or Fable access on Free.
Claude	Pro	$20	Opus, Sonnet and Haiku; Fable is listed under usage credits rather than the included allowance.
Claude	Max 5x	$100	Opus, Sonnet and Haiku; Fable at 50% of weekly limits, tracked as a separate weekly Fable limit.
Claude	Max 20x	$200	Opus, Sonnet and Haiku; Fable at 50% of weekly limits, tracked as a separate weekly Fable limit.
Gemini	Free	Free	Flash-Lite ("Access to our Flash-Lite model"); the page lists no Flash, Pro or Gemini 4 Argon access on Free.
Gemini	Google AI Plus	$4.99	Flash ("Access to our Flash model"); the page lists Gemini 4 Argon only on Google AI Pro and Ultra.
Gemini	Google AI Pro	$19.99	Gemini 4 Argon, the Pro model and the Flash model.
Gemini	Google AI Ultra 5x	$99.99	Highest access to Gemini 4 Argon, the Pro model and the Flash model.
Gemini	Google AI Ultra 20x	$199.99	Highest access to Gemini 4 Argon, the Pro model and the Flash model.

Three lines in that table change the advice above. On ChatGPT Plus, OpenAI lists GPT-6.1 Sol in Work and Codex rather than in Chat, so the chat window isn't where Plus gives you that model. On Claude, Opus starts at Pro, so a free user is testing Sonnet. Fable, which leads most of our rows on the November 10 data, isn't in Pro's included allowance: Anthropic lists it under usage credits on Pro and under a separate weekly limit on Max. On Gemini, Argon starts at Google AI Pro, and the cheaper Google AI Plus lists only the Flash model.

Google sets plan prices by country, and its plan pages show each reader the prices for their own region. The Gemini rows come from Google's United States subscriptions page; outside the US, the Google AI plans page shows your price.

For a single-app decision, Claude plans in detail, Claude Pro vs Max, and whether ChatGPT Plus is worth it go further. If you call these models through an API rather than an app, per-token rates are on LLM pricing and the full model ranking is on the leaderboard.

Microsoft Copilot has no score, so compare it on plans and features

Many readers already pay for Copilot through Microsoft 365, and it's missing from every table above. Microsoft's transparency note says Copilot draws on models from Microsoft, OpenAI, and other developers, and it doesn't say which one answers a given chat. A score for any one of those models wouldn't describe Copilot, so we leave it unscored.

What all four makers do publish is a plan page. This table reads each one for the same features.

Table 5 Download CSV

Plans and features	ChatGPT	Claude	Gemini	Microsoft Copilot
Free plan	Yes.	Yes.	Yes.	Yes.
Cheapest paid plan	Go, $8 a month.	Pro, $20 a month.	Google AI Plus, $4.99 a month.	Microsoft 365 Personal, $9.99 a month.
Plan nearest $20	Plus, $20 a month.	Pro, $20 a month, or $17 a month billed yearly.	Google AI Pro, $19.99 a month.	Microsoft 365 Premium, $19.99 a month.
Models the plan names	GPT-6, with “advanced reasoning models” on Plus.	Opus, Sonnet, and Haiku on Pro.	Gemini 4 Argon on Google AI Pro.	None named. Microsoft says Copilot uses “models from Microsoft, OpenAI, and other model developers”.
Deep research	Free, with limits. Expanded on Plus.	From Pro. The Free column says no.	Free.	From Microsoft 365 Premium, as limited “advanced research and analytics”.
Voice chat	Free, with limits.	Free.	Free, as Gemini Live.	Not named. Paid plans list AI-generated audio (English only) and help from your screen or camera.
Image creation	Free, with limits.	Not listed on the plan page.	Free, with editing.	Free. Paid plans raise the limit.
Memory across chats	Free, with limits.	Free.	Not on the plan page. Google’s privacy hub describes a memory summary you can turn off.	Not on the plan page. Microsoft’s privacy page describes Copilot Memory, which you can turn off.
Projects	Listed on Plus, with scheduled tasks and custom GPTs.	Free, up to five projects.	Not listed. Free includes Gems and Canvas.	Not listed on the plan page.
Scheduled tasks	Listed on Plus.	From Pro.	Not listed below Ultra. Ultra lists first access to “Tasks in Gemini”, in some countries.	Not listed on the plan page.
Coding agent	Codex: limited on Free, expanded on Plus.	Claude Code, from Pro.	Google Antigravity, from Google AI Pro at entry rate limits.	Not listed. GitHub Copilot is a separate product.
Inside office apps	Extensions for Excel, Word, PowerPoint, and Google Sheets, limited on the lower plans.	Claude for Microsoft 365 and Outlook, from Pro.	Gmail and Vids from Google AI Plus. Docs from Google AI Pro.	Word, Excel, PowerPoint, Outlook, and OneNote, from Microsoft 365 Personal.
Cloud storage included	None listed.	None listed.	15 GB on Free, 400 GB on Google AI Plus, 5 TB on Google AI Pro.	1 TB on Microsoft 365 Personal, 2 TB on Premium.

Each cell is what the maker’s own plan page lists and the lowest plan it lists it on. “Not listed” means the page doesn’t mention the feature, which isn’t proof the app lacks it. Plan prices were read October 16, 2026 for ChatGPT and Claude, October 21, 2026 for Gemini, and November 10, 2026 for Copilot.

Sources, read November 10, 2026: ChatGPT pricing; Claude pricing; Google AI plans, Gemini Apps privacy hub; Microsoft Copilot plans for individuals, Microsoft Copilot transparency note, Microsoft Copilot: your privacy controls.

For most people the office row decides it. Copilot is built into the desktop Word, Excel, PowerPoint, and Outlook apps from Microsoft 365 Personal, which also includes cloud storage. ChatGPT and Claude reach those apps through extensions, and Gemini works inside Google's own apps. ChatGPT vs Copilot compares those two plan by plan.

The four apps treat your chats differently

The defaults matter here, because most people never open the settings.

Table 6 Download CSV

Your chats	ChatGPT	Claude	Gemini	Microsoft Copilot
Do your chats train the models?	Yes by default on personal plans.	If you allow it in privacy settings, or if a chat is flagged for safety review. The plan page labels the setting “opt-out”.	Yes while Keep Activity is on, and human reviewers read a subset of chats.	No. Microsoft says prompts, responses, and file contents “aren’t used to train foundation models”.
How to stop it	Turn off “Improve the model for everyone” under Data controls. Voice mode and Codex have their own settings.	Turn off model improvement in Privacy Settings.	Turn off Keep Activity. Chats already reviewed are kept for up to three years.	Nothing to turn off for training. Personalization and memory have their own switches.
Chats kept out of training	Temporary chats. They may be kept up to 30 days for safety.	Incognito chats, even when model improvement is on.	Temporary chats. They are kept with your account for 72 hours.	All chats, on Microsoft’s statement above.
Ads	The Go plan “may include ads”.	Not mentioned on the pages we read.	Google says chats “are not being used to show you ads”.	Possible without a Microsoft 365 subscription. Ad personalization can be turned off.

These terms cover personal accounts. Work, school, and API accounts run under different terms, and each maker can change a default at any time, so open the source before you paste in anything sensitive.

Sources, read November 10, 2026: OpenAI: data controls in ChatGPT, ChatGPT pricing; Anthropic privacy center, Claude pricing; Gemini Apps privacy hub; Microsoft Copilot: your activity history, Microsoft Copilot: your privacy controls.

On personal accounts, ChatGPT and Gemini use your chats for training until you change a setting, and on Claude it depends on a privacy setting. Microsoft says Copilot chats aren't used to train foundation models. ChatGPT, Claude, and Gemini each have a chat mode that stays out of training, and that's the one to use for anything sensitive.

Where each pick is the wrong one

Claude is the wrong first pick if you need the cheapest paid tier or you keep running into usage limits. Its first paid plan costs $20, against $8 for ChatGPT Go and $4.99 for Google AI Plus, and the five-hour session limit is the one we'd check first against your own use.

ChatGPT is the wrong pick for writing-heavy work, where its best model trails on blind creative-writing votes. It's also the wrong pick for daily coding on Plus if the Codex allowance runs out before your week does.

Gemini is the wrong pick when you want the leader on coding or agent tasks, and its top model needs Google AI Pro or Ultra.

Copilot is the wrong pick if you want to know which model is answering. Microsoft doesn't name it, so none of the scores in this post apply to it.

So start with the free plan of the app that leads your main job in the first table. Run the same three real tasks through it and one other app, and pay for the one whose answers you edit least. The tables here re-render with every ranking update, so the row to recheck before you renew is the one for your job.