Claude Free vs ChatGPT Free for Writing in 2026

By Austin Little

You do not need to pay anybody to get a rough draft moving. You need to know what each free tier actually gives you, where it runs out, and how to work so that running out is a coffee break instead of a crisis.

AI disclosure. This page was drafted with AI assistance and edited for Apiary. We don't invent quotes, stats, people, or events. Plan details below were read off the official Claude and ChatGPT pricing pages on October 1, 2026. Both companies change plans often, so re-check before you rely on any single line.

Why this comparison, and why free

Most "which AI is best for writing" articles are really "which subscription should you buy" articles wearing a disguise. They test the paid models, rank them, and then mention the free tier in a footnote. That is backwards for most people who write. Students, nonprofit staff, freelancers between clients, retirees writing family history, people drafting cover letters at the kitchen table — a lot of those folks are not going to spend money every month on a chat box, and they should not be made to feel behind because of it.

So this guide only looks at the free plans. Claude Free from Anthropic. ChatGPT Free from OpenAI. What is on the box, what is not, and how a working writer can get real drafts out of each without hitting a wall at the worst moment.

One promise up front: I am not going to tell you that one of them "wins." They are different tools with different shapes. The useful question is which shape fits the writing you do this week.

What we checked, and how

On October 1, 2026, I read the plan-comparison tables on both official pricing pages: claude.com/pricing and chatgpt.com/pricing. Everything in the side-by-side below comes from those two pages on that day. Where a page says something vague — "limited," "varies," "subject to change" — I repeat the vague wording rather than guessing a number.

What I did not do: run a benchmark, count messages until I got cut off, or claim one model writes "better prose." Those tests go stale in weeks, and a single person's taste in prose is not a fact. You will get more out of a fifteen-minute test with your own writing than out of anybody's ranking, including mine. There is a test plan for that near the end.

The side-by-side, from the official pages

Claude Free, as listed on October 1, 2026

The Claude pricing page lists the Free plan at $0, "free for everyone," with these headline features:

• Chat on web, desktop, and mobile.
• Search the web, create files, and run code.
• Memory across conversations.
• Connect your apps and tools (connectors).
• Create Artifacts.

The detailed comparison table adds a few things writers care about:

• Models: Sonnet and Haiku are marked "Yes" on Free. Opus is marked "No" on Free.
• Projects: "Up to 5" on Free.
• Research (the longer, multi-step research feature): "No" on Free.
• Voice mode, incognito chats, user preferences, skills: all "Yes" on Free.
• Model training: listed as "Opt-out" on every consumer plan, including Free.

On usage, the Claude FAQ says every plan has usage limits that reset on a rolling five-hour session window, that how much you can do depends on conversation length and complexity, the model, and the features you use, and that "there's no fixed message count." It also says "Free covers everyday questions" and that the next plan up gives "at least 5x more usage per 5-hour session than Free." When you hit a limit, the stated options are to wait for the reset or move to a higher plan.

ChatGPT Free, as listed on October 1, 2026

The ChatGPT pricing page describes Free as "Intelligence for everyday tasks" and lists:

• Unlimited text chats with GPT-5.6 Luna (the page footnotes this as "subject to abuse guardrails").
• Limited messages with uploads.
• Limited image creation.
• Limited voice chats.
• Limited deep research.
• Limited memory and context.
• Limited Codex access, and limited ChatGPT Work access on desktop.

The detailed table adds:

• Models: GPT-5.6 Luna and GPT-5 Thinking Mini are marked "Yes" on Free. The newer GPT-6 family models listed on the page are marked "No" on Free.
• Context window for the "Instant" model: 27K total on Free, with a user-input maximum the page approximates as "~12 pages of text." The next plan up lists 54K and about 40 pages.
• Reasoning model context on Free: "Varies."
• Projects: "Yes" on Free. Shared projects: "Yes."
• File uploads, data analysis, vision: "Limited."
• Search: "Yes." Study mode: "Yes."
• Scheduled tasks: "No" on Free.
• Response times: "Limited on bandwidth & availability."
• Training: "Content is used to train our models — Opt-out available."

The page also notes that the stated input space is approximate, because the window is shared with system instructions, memories, and internal reasoning.

Side-by-side table for writers

Read that table as a snapshot, not a contract. Both pages say plans and limits can change.

What those differences mean when you are actually writing

Numbers on a pricing page only matter once you know what a writing session looks like. Here is how the differences show up in practice, in plain terms.

"How much can it hold?" is the question that matters most for long drafts

Context is the amount of text the model can keep in view at once: your instructions, the pasted draft, its replies, and everything else in that conversation. When a conversation gets longer than the window, older material falls out of view, and the model starts forgetting what you told it at the top.

On ChatGPT Free, the official table puts the default model's input space at about twelve pages of text. That is plenty for a cover letter, a blog post, a scene, a chapter section, or a grant narrative paragraph. It is tight for "here is my whole 40-page manuscript, find the continuity errors." If you paste a long document and the answers start ignoring the beginning, that is probably the window, not you.

On Claude Free, the table language is broader, but it says "varies by model," and I am not going to promise you a page count I could not confirm for the free plan specifically. In practice, test it: paste a long document, ask a question that can only be answered from the first page, and see if it gets it right.

Either way, the working habit is the same and it is free: feed long work in sections, with a short summary of the whole at the top of each new conversation. You will get better edits that way on any tool, at any price.

"How often can I use it?" works differently on each

Claude Free describes a rolling five-hour session window with no fixed message count. Long conversations, big pastes, and heavier features draw down faster. That means a writer who pastes a 5,000-word chapter ten times in a row will hit the limit sooner than someone asking short questions. The page does not publish a number, so neither will I.

ChatGPT Free describes text chats on its default model as "unlimited," subject to abuse guardrails, while uploads, images, voice, and deep research are "limited." For a writer who mostly types and pastes text, that default model is the workhorse.

The writer's takeaway: on Claude Free, keep conversations lean and start fresh ones often. On ChatGPT Free, prefer pasting text over uploading files when you can.

Projects: keeping a book or a client separate

Both free tiers now list Projects. Claude Free caps it at five. ChatGPT Free just says "Yes." For a writer, a project is a labeled room: one for the novel, one for the newsletter, one for the grant, one for the day job. You can keep instructions ("my style: short sentences, no exclamation points, American spelling") in one place instead of retyping them.

Five is enough for most people if you are disciplined. Name them by the work, not the date. When a project is done, export what you need and archive or delete the project to make room.

Memory: helpful and a little risky

Claude Free lists memory as "Yes." ChatGPT Free lists it as "Limited." Memory means the tool can carry facts about you between conversations. For writers this cuts both ways. It is nice not to re-explain that you write middle-grade fiction. It is less nice when a client's confidential details from last Tuesday bleed into a different client's draft today.

Simple rule: if you are writing for more than one client, or anything sensitive, check what the memory has saved and use separate projects or an incognito / temporary chat for the sensitive work.

Training on your words

Both pages list training as opt-out on the free plan. That means your chats may be used to improve their models unless you turn that off in settings. If you are writing anything you would not want in someone else's dataset — an unpublished manuscript, a client's internal memo, a personal essay about your family — take two minutes and find the setting before you paste.

This is not paranoia. It is the same habit as not leaving a manuscript on a café table.

Web search and research

Both free tiers list web search. That is useful for writers who need a quick check on a date, a spelling, or a source to read. It is not the same as fact-checking. A search-backed answer can still summarize a source badly or pick a weak source. Click through and read the page yourself before a fact goes into your draft.

Claude Free does not include the longer Research mode. ChatGPT Free lists deep research as "Limited." If your writing depends on assembling sources, do the source-gathering yourself — library databases, the publication's own site, a primary document — and use the AI to help you organize notes you have already verified.

Which free tier fits which kind of writing

None of this is a ranking. These are starting suggestions based on the published plan shapes. Your own test beats my suggestion every time.

Short-form, high-volume: emails, posts, product blurbs

If you write a lot of short pieces in a day, a tier that describes text chat as unlimited on its default model is a natural fit. Short pastes stay well inside a twelve-page window. You can open a fresh chat per piece and never think about limits.

Claude Free also handles short pieces well; just know that a long day of rapid back-and-forth draws on a rolling window. Batch your requests: "Here are five product blurbs, tighten each to 40 words" in one message beats five separate messages.

Long-form: chapters, reports, long essays

Long work is where the context difference matters. Before you commit a manuscript to either tool, run the "first-page test" described below. Whichever tool keeps the beginning in view on your actual document is the one to use for structural edits.

For both tools, the section-by-section method is still the most reliable. Paste a one-paragraph synopsis and the section you are editing. Ask for specific things. Move on.

Editing someone else's work, or your own, line by line

For line edits, what you want is a tool that follows instructions about what not to change. Both can. The trick is in the prompt: "Do not rewrite. Mark only grammar errors and unclear sentences. Return a list, not a new version." Test that exact instruction on both and see which one respects it on your text.

Research-heavy nonfiction

Neither free tier should be your fact source. Use them to outline, to suggest questions you should be asking, to turn your verified notes into a draft, and to flag claims that sound like they need a citation. Then go verify.

Fiction and voice-heavy work

Taste matters most here, and taste is not something a pricing page tells you. Paste the same paragraph of your fiction into both, ask each for "three ways to make this paragraph tenser without changing the events," and see which suggestions sound like something you would actually write. That is your answer, for this book, this month.

A fifteen-minute test you can run today

Here is a fair, free test. Do it once with each tool, with the same inputs.

Step 1: Pick three real pieces

Use your own writing. One short (an email or a 200-word post), one medium (a 1,000-word article or scene), one long (the longest document you regularly work on, broken into its natural sections if needed). Remove anything confidential first.

Step 2: Give both tools the same instructions

Write your instructions once in a plain text file so you paste the identical words into both:

• Your audience and purpose, in one sentence.
• Your style rules (sentence length, banned words, spelling convention, tone).
• What you want back (a list of edits, a revised version, or questions).
• What not to do ("Do not invent facts. Do not add quotes. Mark anything you are unsure of.").

Step 3: Run three checks on each

1. The short-piece check. Ask for a tightened version that keeps your voice. Does it sound like you, or like a brochure?
2. The instruction check. On the medium piece, ask for line edits only, as a list. Did it obey, or did it rewrite the whole thing?
3. The first-page check. On the long piece, paste as much as you comfortably can, then ask a question that can only be answered from the very beginning. Did it remember?

Step 4: Write down what happened

Three lines per tool. Which check it passed, which it failed, and anything that annoyed you. That note is worth more than any review you will read, because it is about your writing.

Working habits that make any free tier enough

The real secret of free-tier writing is not picking the right tool. It is working in a way that does not depend on unlimited anything.

Keep your instructions outside the chat

Write a short "style card" in a plain text file. Paste it at the top of every new conversation. You will never lose your setup to a reset, a limit, or a forgotten memory. This also makes switching between Claude Free and ChatGPT Free painless; the card works in both.

Start fresh conversations on purpose

Long conversations do two bad things: they eat into usage faster, and they push your early instructions out of the model's view. When a conversation has gone more than a few rounds on a long draft, start a new one with your style card, a one-paragraph summary, and the current section.

Paste text instead of uploading files

On ChatGPT Free, uploads are listed as limited; plain text chat is not. On either tool, pasted text is also easier to control: you know exactly what the model is seeing. Copy the section you are working on out of your document and paste it.

Save your good outputs locally, immediately

Do not treat the chat history as your filing cabinet. When you get a paragraph you like, copy it into your own document right away, with a note of which tool and which date. Plans change. Chat histories get cleared. Your folder is yours. (We have a separate Apiary guide on keeping a local writing folder organized.)

Ask for questions before answers

One of the best free-tier moves is to ask, "Before you edit, ask me five questions about what this piece needs." The questions take fewer tokens than a full rewrite, and they often show you the real problem faster.

Have a no-AI fallback

When you hit a limit mid-session, that is not an emergency. Switch to the other free tool with your style card, or close the laptop and do a read-aloud pass. Reading your draft out loud catches more rhythm problems than any model.

Honesty rules that apply on both tools

Whatever free tier you use, a few rules keep your writing trustworthy.

Never let it invent facts, quotes, or sources

Both tools can produce confident sentences that are false, and citations that do not exist. If a draft includes a statistic, a quote, a date, or a source, you check it against the original before it stays. If you cannot find the original, the line comes out.

Add this to your style card: "Do not invent facts, quotes, statistics, names, or citations. If something would need a source, write [SOURCE NEEDED] instead."

Know the disclosure rules where you publish

Some publications, schools, employers, contests, and funders have rules about AI use. Those rules differ widely and they change. Read the actual policy of the place you are submitting to. Do not assume. If there is no written policy, ask the editor or instructor directly.

Keep the private stuff private

Do not paste passwords, account numbers, medical records, or other people's private information into any chat tool, free or paid. For confidential client work, check the training setting and use a separate project or private chat. If the work is truly sensitive, a local model on your own machine is the safer free option — that is covered in our guide to using AI without paying.

When does upgrading actually make sense?

This is a free-tools page, and nothing in it requires paying. But "upgrade panic" — the feeling that you are doing it wrong because a banner says you have run out — deserves a straight answer.

A paid plan makes sense only when you can name a specific job that keeps failing on the free tier and that the paid plan's own page says it fixes. "I hit the limit sometimes" is not that job; batching, fresh conversations, and switching tools usually handle it. "I edit a long manuscript every day and the free window cannot hold the sections I need" might be. Even then, a local model, a free alternative, or a change in method may solve it first.

If a banner makes you feel rushed, close the tab and come back in an hour. Urgency is a sales technique, not a writing need.

A note on the names

Both companies rename models often. On October 1, 2026, the Claude page listed Sonnet and Haiku on Free, and the ChatGPT page listed GPT-5.6 Luna and GPT-5 Thinking Mini on Free. By the time you read this, those names may be different.

Quick answers

Is Claude Free actually free?

Yes. The official page lists the Free plan at $0, "free for everyone," with usage limits that reset on a rolling five-hour window.

Is ChatGPT Free actually free?

Yes. The official page says the free version is available to everyone and lists unlimited text chats on its default model, subject to abuse guardrails, with limits on uploads, images, voice, and deep research.

Which one is better for writing?

That depends on your writing. Run the fifteen-minute test with your own pieces. For long documents, the first-page check is the deciding test.

Can I use both?

Yes, and many writers do. Keep a style card in a text file and paste it into whichever tool is available. When one hits a limit, switch.

Do I need to pay to write a book with AI help?

No. Section-by-section work, a style card, and saving outputs locally will carry a book draft on free tiers. Paying is optional, and only worth it for a named job that keeps failing.

Will they train on my writing?

Both list training as opt-out on Free. Find the setting and decide before you paste anything you care about.

The short version

Claude Free and ChatGPT Free are both real, usable writing tools at $0. Claude Free gives you Sonnet and Haiku, up to five projects, memory, and a usage limit that resets on a rolling five-hour window. ChatGPT Free gives you unlimited text chat on its default model, with a smaller published input window of about twelve pages and limits on uploads and heavier features.

Neither one replaces your judgment, your sources, or your voice. Write a style card. Work in sections. Save locally. Verify every fact. And when a banner tells you to upgrade, remember that the draft is yours, not theirs.