ChatGPT Free Plan Limits in 2026
I went looking for the number everyone quotes — ten messages every five hours — and could not find it anywhere in OpenAI's current documentation. What I found instead is that the limit moved. Everyday text chat is now unlimited on the free plan, and the metering has shifted onto the tools: uploads, images, voice, data analysis. Here is the version that matches what a free account actually does in November 2026.
Short answer: on the free plan you can send as many text messages as you like — OpenAI's documentation says "unlimited everyday text chats, subject to abuse-prevention safeguards". What is metered is everything around the text: file uploads, image generation, voice, data analysis and GPTs each have their own separate allowance, and OpenAI does not publish those numbers. Free accounts run on GPT-5.6 Luna and cannot reach GPT-5.6 Sol. When you hit a tool limit you get a notice naming that tool and a reset time — the exact wording is quoted further down.
1. The number everyone quotes is out of date
Search for this question and you will find the same figure repeated across hundreds of pages: free ChatGPT is capped at 10 messages every 5 hours. Some articles add that it drops you to a "mini" model afterwards.
That was true. It was the rule for the GPT-5.5 generation and the models before it. The problem is that OpenAI's article on this subject — the one whose URL ends in 11909943 — has since been rewritten for a different model generation, and the cap is no longer in it.
What the current version says, under the heading How do rate limits work on the Free tier?:
"Free users have unlimited everyday text chats, subject to abuse-prevention safeguards. File uploads, image generation, voice, data analysis, and other tools have separate usage limits. ChatGPT will notify you when you reach an applicable limit."
That is not a paraphrase. It is what the page said when I opened it on 2 November 2026, and I have put a screenshot of it further down so you can check the wording yourself.
This matters for a practical reason. People arrive at this question trying to work out whether they are about to be cut off. The old answer made them ration messages. The current answer is that messages are the one thing you never run out of — the thing that will stop you is an upload, an image, or a very long conversation.
2. Proof: what a free account looks like
Documentation and a live account are two different things, so here is the account. It has never had a payment method attached, and the sidebar says so plainly.
Four things in that screenshot are worth reading, because each one is a limit in disguise:
- The plan label says "Free". Not a trial, not a downgraded Plus. This is the tier being described.
- The "Claim offer" button is present. That is the upgrade prompt, and it is the same one you will see again inside a limit notice.
- The composer has a "Think" toggle. Free users do get thinking — which surprises people who remember when they did not.
- Images, Scheduled, Library, Plugins, Codex and Explore are all in the sidebar. Every one of those is a metered surface. The text box is the only unmetered thing on the screen.
3. What the free plan actually limits
Reading the documentation and the account together, this is the shape of the free tier as it stands:
| Surface | What free gets | Is a number published? |
|---|---|---|
| Text chat | Unlimited, subject to abuse-prevention safeguards | No number to publish |
| File and image uploads | Separate allowance | No |
| Image generation | Separate allowance | No |
| Voice | Separate allowance | No |
| Data analysis | Separate allowance | No |
| GPTs (using them) | Follows your free model limit; access pauses at the cap | Same as the model |
| Creating GPTs | Not available on any personal account, Free through Pro | — |
| Library storage | 500 MB | Yes — the one hard number |
Note how little of that column is a number. 500 MB of Library storage is the only fixed quantity OpenAI publishes for the free tier. Everything else is described as "separate limits" without a figure attached, and the documentation is explicit that the notice appears when you reach one rather than beforehand:
"If you reach a rate limit for any of these tools, you will need to wait for a later time to use any of these tools."
This is why every "ChatGPT free limits" article that lists precise numbers for uploads or images is guessing, extrapolating, or quoting an old build. If a page tells you free gets exactly three image generations a day, ask where the number came from, because it is not on OpenAI's site.
4. What the notice says, word for word
The documentation shows the notice as an illustration, and it is worth reading closely because it tells you three useful things at once:
The notice reads:
"You've reached your image creation limit."
"Upgrade to ChatGPT Plus or try again tomorrow after 9:17 AM."
Three details worth extracting:
- It names the surface. "Image creation limit" — not "your limit". Each tool has its own counter, so the notice tells you which one you drained.
- It gives an exact reset time. "Tomorrow after 9:17 AM", not "in 24 hours". The documentation elsewhere says ChatGPT "shows the reset timing in-product", and this is what that looks like.
- It offers the upgrade inline. The Get Plus button sits inside the notice, which is why so many people describe this as a paywall rather than a quota.
5. Luna vs Sol: what $0 does not include
The other limit is the one nobody calls a limit: model access. Free accounts get GPT-5.6 Luna. They do not get GPT-5.6 Sol, and the availability table in OpenAI's documentation marks it plainly.
| Thinking level | Plus | Pro | Business | Free & Go |
|---|---|---|---|---|
| Medium / High (GPT-5.6 Sol) | Included | Included | Included | Not included |
| Extra High | Not included | Included | Included | Not included |
| Pro | Not included | Included | Included | Not included |
The subtlety that catches people out is the Think button. It is on the free composer, and it does change behaviour — but it does not upgrade the model:
"Free and Go users can use Think for harder questions. Think uses GPT-5.6 Luna, not GPT-5.6 Sol."
So pressing Think on free gets you more time on the same model, not a better model. That is a real difference for a hard maths problem and no difference at all for a hard coding problem, where the ceiling is the model itself rather than how long it is allowed to think.
Logged-out users get neither. The documentation is explicit that signing out removes Sol access entirely, which is worth knowing if you have been comparing ChatGPT to something else using a logged-out browser.
6. Three different walls people call "limits"
Most confusion about this topic comes from three unrelated things sharing one word. They have different causes and different fixes.
Wall one: the tool allowance
What section 4 describes. Uploads, images, voice, analysis. Fix: wait for the reset time in the notice, or upgrade. Nothing you can do in settings.
Wall two: the context window
The message "Context window full" or "Chat memory full". This is per conversation, not per account. Your messages and the replies both occupy it, which is why a few long exchanges fill it faster than many short ones.
The documentation's own advice is blunt: there is no fixed number of messages per chat, and the remedy is to start a new chat — which free users can do immediately. You do not need to turn off Memory, delete saved memories, or clear other chats. A full chat costs you the thread's continuity, nothing else.
Wall three: the abuse safeguard
"Subject to abuse-prevention safeguards" is the phrase doing quiet work in the unlimited promise. It is not a published quota and it is not something a normal person will meet. It exists so that automated scraping cannot use the free tier as a free API. If you are sending thousands of requests programmatically and something throttles you, this is the clause you are running into — and the fix is the API, not a different account.
7. Working out which wall you hit
Four signals, in the order that resolves it fastest:
- The notice names a tool (image creation, uploads, data analysis) → wall one. Note the reset time, wait, or upgrade.
- The notice says "context window full" → wall two. Start a new chat. Same account, no penalty.
- No notice, but the answers got worse → you are probably on Luna and expected Sol, or on a GPT whose own cap has been reached and which has dropped to a smaller variant. Check which model the selector shows.
- No notice, requests just stop completing in an automated script → wall three. Move to the API.
One more signal that misleads people: the "Claim offer" button is not a limit indicator. It sits on the account row permanently as an upgrade prompt, including on accounts that have never hit anything. If you have been reading its presence as "I am near a limit", it is not.
8. What Plus actually changes
Given that the free plan's text chat is unlimited, what is the $20 buying? Not message count. It buys:
- Model access. GPT-5.6 Sol at Medium and High thinking levels, which free never gets. This is the substantive difference.
- Larger tool allowances. The upload, image, voice and analysis counters are bigger — with numbers that are not published either.
- A different peak behaviour. Paid plans get their own quota windows and, per the documentation, a weekly cap on the top thinking tier rather than a per-conversation one.
If your reason for considering Plus is "I keep running out of messages", check that first — on the current free plan that is no longer a thing that happens to text chat. If your reason is "the answers are not good enough on hard problems", that one is real, and it is the Sol gap.
We compared the three big free tiers on the same prompts in DeepSeek vs ChatGPT vs Gemini, and the pattern there is the same one described here: the free tiers differ far more in which model answers you than in how many messages they allow.
Frequently asked questions
How many messages can I send on ChatGPT free per day?
There is no daily number any more. OpenAI's documentation says free users have unlimited everyday text chats, subject to abuse-prevention safeguards. The tools — uploads, images, voice, data analysis, GPTs — each carry a separate allowance, and those numbers are not published.
Is the 10 messages every 5 hours limit still true?
It was accurate for GPT-5.5 and earlier, and a lot of pages still quote it. The same OpenAI article now describes GPT-5.6 Luna as the free model and says text chat is unlimited. If a guide gives a hard number, check its date and the model it names.
What model do free users get?
GPT-5.6 Luna. Free and Go users do not get GPT-5.6 Sol; Medium, High, Extra High and Pro are all marked "not included" for them. Think is available on free, but it runs Luna as well — more thinking time, same model.
What happens when you hit a limit?
A notice naming that specific limit, with a reset time and an upgrade button. The wording we captured: "You've reached your image creation limit. Upgrade to ChatGPT Plus or try again tomorrow after 9:17 AM." Text chat does not produce this notice.
Does "context window full" mean I used up my messages?
No. It is a per-conversation size limit, not a plan allowance. Start a new chat — free users can do it immediately, and you do not need to delete memories or clear other chats.
Sources and what we could not test
The quoted sentences come from OpenAI's help centre, read on 2 November 2026: the ChatGPT Free Tier FAQ for the rate-limit answer and the separate-tools list, and the GPT-5.6 and GPT-6 Pro article for the Luna/Sol availability table. Both were checked in more than one language on the same day and said the same thing.
| Claim | Where it comes from |
|---|---|
| Free text chat is unlimited | OpenAI help centre, read 2 Nov 2026 — quoted verbatim |
| Uploads / images / voice / analysis have separate limits | Same article, plus its list of tools with separate limits |
| Free gets GPT-5.6 Luna, not Sol | Availability table in the GPT-5.6 article |
| Think on free runs Luna | Same article, quoted verbatim |
| Library storage is 500 MB | Free Tier FAQ |
| The limit notice wording | Screenshot of OpenAI's article, captured 2 Nov 2026 |
| The plan label on a real account | Our own free account, screenshot above |
What we could not test: the actual size of the upload, image, voice and analysis allowances — OpenAI does not publish them, and we did not burn through them to find the edges, because that would have meant deliberately blocking our own account for a day. Whether the "unlimited" text chat has a hidden threshold in practice: we sent a normal volume of messages, not thousands, so we can confirm it is not the old 10-per-5-hours cap but we cannot say it is truly infinite. Whether every country sees the same plan: the documentation notes that ads appear on the free tier in some countries, so regional builds already differ. And how any of this behaves on the mobile app, which is where Think is most fully rolled out.
One thing we deliberately did not do: repeat the old numbers to look thorough. Every article that tells you free gets exactly ten messages every five hours is now wrong, and adding one more of them would make the problem worse.