Is ChatGPT Actually Free in 2026? The Real Limits
Yes, ChatGPT is still free in 2026. No card, no trial clock — and as of the week of 10 August, freer than it used to be: OpenAI’s help pages now state, in present tense, that Free users have unlimited everyday text chats. The paid side moved in September. Pro is now three tiers, topping out at $500, and you can’t build a new custom GPT on any personal plan anymore, paid or not. Custom GPTs themselves are being retired on 11 December. And since 8 October GPT-6 Luna has been rolling out as the free main chat model, per OpenAI’s release notes, even though its Free Tier FAQ still names the older GPT-5.6 Luna.
Last verified: 31 October 2026. Plan prices (except Go’s $8, last confirmed 3 October), the Pro tiers, Free limits, file caps, voice allowances, the custom GPT retirement, context windows, model access, Business seat prices and the Business per-seat message ranges were re-checked against OpenAI’s help pages, release notes and pricing page on that date. Library storage and the search, memory and deep research rows were last confirmed on 3 October and weren’t re-checked this pass. Go having no annual billing comes from the August wording of the Pro tiers page, and it’s marked as such in the table. The Go Think cap this page carried in August (300 messages per 3 hours) is gone from OpenAI’s pages, so it’s gone from the table too. Where OpenAI publishes no number, this page says “not published” instead of borrowing one from somebody else’s blog.
What changed on 6 August
OpenAI posted on 6 August 2026 that “GPT‑5.6 Luna will become the default model for Free and Go users this week.” The help docs say the same thing and add that Free and Go users don’t get GPT-5.6 Sol at all; Luna is their default, and Luna is also what powers the Think feature.
The swap wasn’t clean. On 8 August, OpenAI’s GPT-5.6 help page said in one section that Luna was “becoming the default model for Free and Go users” and in another that GPT-5.5 Instant “remains the default model.” By 11 August both stale lines were gone, so if your free account hit a wall that week, the change was still in flight and your account wasn’t broken.
The bigger part landed on schedule. The post promised: “Starting next week, they’ll also have unlimited text chats and access to a new Think button for harder questions (subject to abuse guardrails)” — and in the week beginning Monday 10 August, OpenAI’s help pages flipped to present tense. The Free Tier FAQ now reads: “Free users have unlimited everyday text chats, subject to abuse-prevention safeguards.” The GPT-5.6 page said the same for Go in August, and documented Think as live: “Free and Go users can use Think for harder questions in ChatGPT on the web and in the mobile app.” The October version of the Free Tier FAQ is more cautious about the web half: “Think is available in the ChatGPT mobile app and is rolling out on the web.” So if the Think button is missing in your browser, try the phone app. One oddity from August: the announcement post itself was never updated, and still promised in future tense what the help pages documented as fact.
Unlimited is OpenAI’s word, and the very next sentence in OpenAI’s own post walks it back to the part that matters: “Limits will still apply for file uploads, images and other tools.” So the honest reading is that plain back-and-forth typing stops being metered, while everything that costs real compute — attaching a PDF, generating an image, running data analysis — stays capped.
OpenAI has published some of those caps already, on a page most write-ups miss: the File Uploads FAQ limits Free users to 3 file uploads per day, against an “up to 80 files every 3 hours” ceiling that applies across plans. Those caps survived the change; the File Uploads FAQ was untouched by the rollout.
The GPT-5.6 Luna default didn’t last. It was replaced in October, and that’s covered below. The Free Tier FAQ hasn’t caught up, though. It still says “Free users have access to GPT-5.6 Luna.”
What changed in September
Three things, none of them on the Free tier’s main chat.
A new model generation arrived. GPT-6 Astra came first, on 3 September, rolling out to Plus, Pro, Business and Enterprise. GPT-6 Sol and GPT-6 Luna followed on 22 September, available “in ChatGPT Work and Codex starting today for all Plus, Pro, Business, Enterprise, and Edu users.” Then on 29 September came GPT-6.1 Sol, which the release notes say “improves on GPT-6 Sol in agentic coding, computer use, and professional work.” It’s rolling out to Pro first, then Plus, Business, Enterprise and Edu, so if you’re on Plus and don’t see it yet, that’s why. In September, whether Free and Go got any GPT-6 model depended on which OpenAI page you read. The Sol and Luna announcement said “Free and Go users can access GPT-6 Luna in the desktop app,” but when I checked on 3 October the model table on chatgpt.com/pricing marked GPT-6 Luna “No” for both. Four days later OpenAI made the argument moot.
Pro split into three tiers, Pro 100, Pro 200 and Pro 500, at $100, $200 and $500 a month. More on that below the table.
And custom GPTs started going away. First creation went workspace-only. The Free Tier FAQ now answers the question like this: “New GPT creation and publishing are not available on personal ChatGPT accounts, including Free, Go, Plus, and Pro. You can continue using existing GPTs.” Then the 11 September release notes went further: “We’re planning to retire custom GPTs across ChatGPT plans and provide a migration path to plugins.” The retirement FAQ puts the standard date at 11 December 2026. FedRAMP workspaces and Enterprise workspaces with an approved deferral get until 11 February 2027. New GPT creation in affected Enterprise workspaces ended on 26 October 2026, as scheduled, with FedRAMP workspaces excepted; the freeze doesn’t stop edits to existing, unmigrated GPTs. Meanwhile, “Existing GPTs remain usable until the retirement date.” So if you were planning to pay $8 for Go just to build a GPT, which is what this page used to suggest, don’t. Same goes for buying Business seats to get around it. You’d be building on something with under six weeks left.
What changed on 7 and 8 October
GPT-6 moved into the main chat, Free tier included. OpenAI’s 7 October release notes: “The rollout starts globally today for Plus, Pro, Business, and Enterprise in the Chat tab, and expands to Free and Go starting tomorrow… Plus, Pro, Business, and Enterprise use GPT-6 Sol; Free and Go use GPT-6 Luna.”
So from 8 October, GPT-6 Luna has been rolling out as the main chat model on free accounts. The pricing page agrees. The Free plan card now says “Unlimited text chats with GPT-6,” and the model table marks GPT-5.6 Luna “No” for Free and for Go.
Two other OpenAI pages didn’t get the memo. The Free Tier FAQ still says “Free users have access to GPT-5.6 Luna,” and the GPT-6 models help page says “Free and Go users receive GPT-5.6 Luna,” three weeks after the rollout began. I’m going with the release notes and the pricing page, because they’re the newer statements and they agree with each other. But if you land on the FAQ and see GPT-5.6, that’s why. This is the second time this year OpenAI’s help centre has contradicted itself about the Free model. In August the stale lines were gone within days. This one was still there when I checked on 31 October.
The verified price and limit ladder
| Free | Go | Plus | Pro | Business | |
|---|---|---|---|---|---|
| Price | $0 | $8/mo (US; localized in some markets; last confirmed 3 Oct) | $20/mo | Pro 100: $100/mo. Pro 200: $200/mo. Pro 500: $500/mo | Standard: $25/user/mo monthly, $20 annual. Premium: $125/user/mo monthly, $100 annual. Min. 2 seats in total |
| What the price buys | — | — | — | Pro 200 has more usage than Pro 100; Pro 500 the most, and it’s the only tier with Ultrafast (GPT-6 Astra and GPT-6.1 Sol) | Per seat; Premium gets higher included usage for Work and Codex and no 5-hour limit |
| Annual billing | — | None (last confirmed Aug) | None | None | Yes, both seat types |
| Main chat model | GPT-6 Luna, rolling out from 8 Oct (release notes, pricing page). Free Tier FAQ and GPT-6 models page still say GPT-5.6 Luna | Same as Free | GPT-6 Sol in the Chat tab since 7 Oct | GPT-6 Sol, as Plus. GPT-6 Pro: the GPT-6 models page lists Pro $100 and $200 only; the Pro tiers page says Pro 100 and Pro 500 keep their weekly GPT-6 Pro limits in Chat | GPT-6 Sol, rolling out to eligible workspaces. Pro messages: Standard 15 per month, Premium 50 per week |
| Text chat cap | Unlimited everyday text chats | Same as Free | Everyday text chats: Unlimited* | Everyday text chats: Unlimited*. Also Unlimited* on GPT-5.6 Sol, Terra, Luna and GPT-5 Thinking Mini | Standard: “Virtually unlimited access applies to Instant chat”, but not to reasoning, Pro, ChatGPT Work, Codex or other tools. Premium: no 5-hour limit |
| Files per project | 5 | 25 | 25 | 40 | 40 |
| Simultaneous uploads (within Projects) | 10 | 10 | 10 | 10 | 10 |
| File uploads | 3 per day | Up to 80 files per 3h | Up to 80 files per 3h | Up to 80 files per 3h | Up to 80 files per 3h |
| Live voice, per rolling 24h | Limited GPT-Live-1.1 Quip (“Limits may change”) | Voice page: 3h GPT-Live-1.1 Quip. Go help page: same limits as Free | 3h GPT-Live-1.1 | Voice page: $100: 15h GPT-Live-1.1. $200: unlimited GPT-Live-1.1. $500: no figure confirmed. Pricing page: “Unlimited*“ | Standard: 3h GPT-Live-1.1, Premium: 15h; then 1.25 credits/min |
| Library storage | 500 MB | 4 GB | 20 GB | 100 GB | 20 GB |
| Web search | Yes | Yes | Yes | Yes | — |
| Memory (and memory with past chats) | Limited | Yes | Expanded | Expanded | — |
| Apps connecting to internal tools | No | No | Yes | Yes | — |
| Deep research | Limited | Limited | Yes | Yes | — |
| Create new custom GPTs | No | No | No | No | Yes, if workspace settings allow. All GPTs retire 11 Dec 2026 |
| Messages per N hours | Everyday text: unlimited | Everyday text: unlimited. No Think cap published as of 3 Oct | No absolute number for reasoning or GPT-6 models | No absolute number for reasoning or GPT-6 models | Standard, per 5h: GPT-6 Astra 5–45; GPT-6 Sol / 6.1 Sol 15–150; GPT-6.1 Terra 35–285; GPT-5.6 Sol 10–100; GPT-5.6 Terra 25–200; GPT-5.6 Luna 250–2,000. Premium: no five-hour limit, and OpenAI’s page now states no multiplier over Standard |
| Images per day | Not published (as of 3 Oct) | Not published (as of 3 Oct) | — | Unlimited and faster image creation | — |
*The pricing page footnote: “Usage must be reasonable and comply with our policies. Usage refers to Chat limits.”
A dash (—) means I didn’t find the figure stated for that plan on the pages I checked, not that the plan lacks the feature.
Two things in that table are worth more than the rest. The first is that you can’t buy a discount by committing to a year on an individual plan. Plus is “$20/month (billed monthly)”, and the Pro tiers page says “Pro is billed monthly. Annual billing and payment for multiple months in advance are not available.” In August that same page named Go alongside Plus and Pro; the October wording I checked only covers Pro, so treat Go’s “none” as August’s answer. Business is the only plan with an annual rate, and it’s the one aimed at teams.
The second is the Pro ladder, which got a lot less tidy. OpenAI used to describe its two Pro tiers as multiples of Plus with the same core capabilities. That framing is gone. The page now just ranks the three against each other: “Pro 200 includes more usage than Pro 100. Pro 500 offers the highest included usage of the three plans.” The pricing page does give every Pro buyer “Unlimited*” everyday text chats, plus “Unlimited*” on GPT-5.6 Sol, Terra, Luna and GPT-5 Thinking Mini. For reasoning and the GPT-6 models, though, there’s no absolute number for any of the three tiers.
What the $500 buys beyond headroom is speed. The Pro tiers page marks Ultrafast “Included” on Pro 500 and “Not included” on Pro 100 and Pro 200, and “At launch, buying credits on Pro 100 or Pro 200 does not unlock Ultrafast.” It covers two models now, not one, and it burns allowance fast: “Fast mode uses your included allowance at 2.5 times the rate, and Ultrafast for GPT-6 Astra and GPT-6.1 Sol uses it at 8 times the rate.” Run out and keep going on credits, and “When using purchased credits, Ultrafast is charged at 6× the Standard credit rate.” I’d keep it for the moments you’re actually waiting on a response.
Whether GPT-6 Pro is on all three tiers depends on which OpenAI page you read. The GPT-6 models help page and the Business models-and-limits page both say “GPT-6 Pro, powered by GPT-6 Astra, is available in ChatGPT for Pro $100, Pro $200, Business, and Enterprise plans.” No Pro $500. The Pro tiers page, though, says “Pro 100 and Pro 500 keep their current weekly GPT-6 Pro message limits in Chat,” which only makes sense if Pro 500 has it. I can’t tell you which page is current. If GPT-6 Pro is the reason you’re looking at Pro 500, check the model picker before you pay.
Voice is still the clearest capability gap between the lower two. The Voice help page gives Pro 100 15 hours of GPT-Live-1.1 against unlimited on Pro 200, and “In Chat mode, Live usage is measured over a rolling 24-hour period,” so it isn’t a midnight reset. Odd detail: chatgpt.com/pricing marks Pro voice “Unlimited*” with no tier split, which doesn’t match the Voice help page. I’d trust the 15-hour figure on Pro 100 until OpenAI lines the two pages up.
If you’re on Pro 200, you can take 29 October out of your calendar. Earlier versions of this page told you to put it there, because OpenAI’s Pro tiers page said the older, larger Pro 200 allowance was only grandfathered until that date, after which you’d pay the same $200 for less. It never happened. The page now reads: “The reduction announced in our September 29 email is canceled. If you already had the higher Work and Codex allowance, you keep it. If you had the lower allowance, it has doubled to match. Your subscription price stays at $200/month.” Pro 200 is open to new subscriptions again, too. And the other two tiers got a bump in the same change: “Pro 100 and Pro 500 now include twice their previous Work and Codex allowance,” which existing subscribers get automatically, without waiting for renewal. Good news for once. It’s also a fair lesson in how much weight a dated cutoff on an OpenAI help page deserves.
Web search is the row people undervalue. The pricing page marks Search “Yes” on Free, Go, Plus and Pro alike, and the Free Tier FAQ lists “Search the web and obtain up-to-date information” outright. You don’t have to pay to get current information out of ChatGPT. Memory is where Free gets visibly trimmed: “Limited” on Free, “Yes” on Go, “Expanded” on Plus and Pro, and the same split for memory with past chats. Apps that connect to internal tools start at Plus.
The storage and file numbers are the sharpest real difference between Free and paid for people doing actual work. Library storage runs 500 MB on Free, 4 GB on Go, 20 GB on Plus and Business, 100 GB on Pro — a 200x spread, and the widest gap on this ladder, between Free and Pro. Projects allow 5 files per project on Free and 40 on Pro and Business. The 10-file simultaneous upload ceiling applies to every plan, Pro 500 included, but note that it’s a Projects limit specifically, not a global one.
In August, OpenAI’s Projects article said 25 files per project for Plus and the File Uploads FAQ said 20. That’s fixed. Both pages now say 25 for Go and Plus.
The numbers everyone quotes that OpenAI does not publish
Search “ChatGPT free tier limit” and you will get a specific message count per 5 hours and a specific number of images per day, repeated across dozens of pages with total confidence. I’m not printing either one, and here’s why.
The message count is moot now that everyday text is unlimited. The image number I still can’t find: not on the Free Tier FAQ, the GPT-5.6 page, the File Uploads FAQ or chatgpt.com/pricing, all checked again on 3 October 2026. (A 2024 release-note archive does mention two DALL·E 3 images a day for Free, but the DALL·E GPT was retired on 30 August 2026 and the note describes a product two years gone, so it is not a current limit.) What OpenAI’s Free Tier FAQ actually says is: “File uploads, image generation, voice, data analysis, and other tools have separate usage limits.”
Separate from what, and how big, it doesn’t say. That’s a deliberate non-answer.
What makes that interesting is that OpenAI is not generally cagey about numbers. Across its help centre it publishes plenty of hard caps, several of which are in the table above and almost none of which get quoted: 3 file uploads per day on Free, 80 files every 3 hours, 500 MB of Library storage on Free against 100 GB on Pro, and 250–2,000 GPT-5.6 Luna messages per 5-hour window on a Standard Business seat. Those are specific, and they’re published.
So the silence is narrow and it’s deliberate. The figures OpenAI declines to state are the ones people actually search for: how many images a free user gets per day, and how much reasoning and GPT-6 usage Plus and Pro subscribers get. Everyday text is covered: the pricing page says “Unlimited*” on every individual plan, footnoted “*Usage must be reasonable and comply with our policies. Usage refers to Chat limits.” Past that, the Pro tiers page only ranks its three tiers against each other, and the GPT-6 models help page says just “Thinking limits depend on your plan.” Neither gives an absolute number for reasoning or GPT-6 usage (checked 31 October). Those allowances move with load, which is likely why OpenAI doesn’t commit to them in writing. A community number that was true for someone in March, in one region, on one account, is not a published limit — and quoting it as one is how bad reference pages get built. I took the same approach when I checked whether the Claude API is free: several vendors have quietly stopped publishing free-tier numbers, and the correct response is to say so.
Two more things that don’t reduce to a number:
Deep Research on Free. The pricing page marks it “Limited” on Free and Go. OpenAI’s FAQ confirms allowances “vary by plan” and points you at an in-product counter, and I don’t have a Free-tier number to give you. The one useful detail it does give applies to everyone: where a plan has a fixed monthly allowance, “it resets every 30 days from the date of your first use.” Not calendar month. If you burned your allowance on the 20th, that’s your reset date now, and it drifts.
Ads. Two OpenAI pages disagree on tense. The ChatGPT Go announcement says OpenAI plans to test ads in the free tier and Go in the US “soon,” which reads as pending. The Free Tier FAQ says ads “may appear in the Free tier in certain countries,” which reads as already possible somewhere. Whether ads are live for you today is not answerable from OpenAI’s published material, and I’m not picking a side.
Two things that are simply gone
People still ask about both, so: Sora is not part of the free ChatGPT product, or the paid one. OpenAI’s own page says “The Sora web and app experiences were discontinued on April 26, 2026,” with the API following on 24 September 2026. And the DALL·E GPT inside ChatGPT was retired on 30 August 2026. If you had image history in it and didn’t download it beforehand, as the release note advised, I’m sorry.
The context window caveat
The per-plan numbers live on chatgpt.com/pricing. “GPT Instant total context window” is 27K on Free, 54K on Go and Plus, 128K on Pro. The “GPT Reasoning” window reads “Varies” on Free, 256K on Go and Plus, 400K on Pro.
27K on Free is small. Paste a long contract or a big log file and the start of the conversation falls out of view sooner than you’d expect. If that’s your workload, Go doubles it for $8 (price last confirmed 3 October), and that alone might be the better reason to pay than anything on the model list.
When paying actually makes sense
If you’re hitting the wall on Free, the one genuinely useful mechanic is that you don’t wait it out: OpenAI’s FAQ says “if you reach a rate limit on your Free tier account, your usage rates will reset when you update your account to a Plus plan, a Pro plan, or a Business plan.” Upgrade mid-block and you’re unblocked immediately.
Beyond that the ladder is reasonably clean. Go at $8 buys you 25 files per project instead of 5 and 4 GB of Library storage instead of 500 MB. Don’t count on voice as a reason to buy it. OpenAI’s Voice help page lists “Go | 3 hours with GPT-Live-1.1 Quip,” but the Go help page says “Voice mode is included with the same usage limits as the Free tier,” and Free only gets “Limited access.” Those can’t both be right. It no longer buys you the GPT creator; nothing personal does, and custom GPTs retire on 11 December regardless. Plus at $20 keeps the same 25-file Projects cap, so the honest case for Plus over Go is model access (GPT-6 Sol in the main chat instead of GPT-6 Luna, plus GPT-6 Astra), the full GPT-Live-1.1 voice model instead of the Quip variant, expanded memory, apps that connect to internal tools, and 20 GB of storage against Go’s 4 GB. Pro starts at $100 with 100 GB of storage and 15 hours of voice, and climbs to $500 for the most usage and Ultrafast.
A Standard Business seat at $20 billed annually lands on exactly the same monthly price as Plus with the same 20 GB, which is worth knowing if there are two of you and you assumed the team plan carried a premium. Don’t buy seats to build custom GPTs, though. The 26 October creation cutoff applied to affected Enterprise workspaces, but OpenAI is retiring custom GPTs across ChatGPT plans on 11 December 2026 and pointing everyone at plugins instead. Two caveats if you’re buying seats. OpenAI’s billing page says “2 seat minimum per workspace. You can mix and match seat types,” so $20 a head means $40 a month, not $20. And Premium seats are no longer a waitlist item; they’re live at $125 per user per month billed monthly or $100 billed annually (both re-confirmed 31 October), with higher included usage for Work and Codex, no 5-hour limit, 15 hours of voice, and 50 Pro messages a week against Standard’s 15 a month.
That 5-hour limit used to be my favourite example of OpenAI confirming a cap exists without saying what it is. In August the Premium announcement promised “no five-hour usage limit” while neither of the two OpenAI pages I’d checked gave Standard seats a number. That’s changed. The Business models-and-limits page now publishes “allowed local messages per 5-hour window” for Standard seats, model by model: 5–45 on GPT-6 Astra, 15–150 on GPT-6 Sol and 6.1 Sol, 35–285 on the newer GPT-6.1 Terra, 10–100 on GPT-5.6 Sol, 25–200 on GPT-5.6 Terra, 250–2,000 on GPT-5.6 Luna. They’re wide ranges, an eightfold spread on Luna, but they’re numbers, and credit where it’s due. Premium is the opposite story. Earlier versions of this page said Premium seats got 5x Standard’s usage, and worked out per-model Premium ranges from that. OpenAI’s page no longer states a multiplier. It now says only that “Premium seats have higher included usage for Work and Codex and no five-hour usage limit,” so I’ve dropped both the 5x and the range built on it. Read the fine print on “unlimited” too: “Virtually unlimited access applies to Instant chat,” and the same page adds that it “does not make reasoning, Pro, ChatGPT Work, Codex, or other tools unlimited.”
For most people asking whether ChatGPT is free, the answer they actually want is “free enough” — and since the August change, for text-only chatting, it plainly is. If your use case is programmatic rather than conversational, that’s a different question with a different answer; my free AI APIs roundup covers the metered side, and the tools I actually use daily piece covers what earns a subscription.
This page was re-verified against OpenAI’s pages on 31 October 2026, two days after the Pro 200 cutoff that OpenAI ended up canceling. Library storage is still as of 3 October. The next date worth watching is 11 December, when custom GPTs retire.
Frequently Asked Questions
Is ChatGPT free in 2026?
Yes. There's still a no-payment Free tier at chatgpt.com, and as of 31 October 2026 OpenAI's Free Tier FAQ says Free users have unlimited everyday text chats, subject to abuse-prevention safeguards. File uploads, image generation, voice, data analysis and other tools have separate limits.
Which model does ChatGPT Free use?
GPT-6 Luna, which OpenAI began rolling out to Free on 8 October 2026, according to its release notes and pricing page. OpenAI's Free Tier FAQ still says GPT-5.6 Luna, so the company's own pages disagree.
How many messages can you send on ChatGPT Free?
For everyday text chats there's no cap to publish: the Free Tier FAQ states unlimited everyday text chats, subject to abuse-prevention safeguards. Tools like file uploads and image generation keep separate limits, and only some of those have published numbers (3 file uploads a day is one).
What does each ChatGPT plan cost?
Go is $8/month in the US (last confirmed 3 October 2026), Plus is $20/month, and Pro comes in three tiers at $100, $200 and $500 a month. A ChatGPT Business Standard seat is $25 per user per month billed monthly or $20 billed annually; a Premium seat is $125 monthly or $100 billed annually. Plus and Pro are billed monthly only.
Can ChatGPT Free users generate images?
Yes, image creation is on the Free plan, with its own usage limit separate from the main chat. As of 3 October 2026, OpenAI didn't publish a daily image number for Free on its Free Tier FAQ, File Uploads FAQ, GPT-5.6 page or pricing page.
Does my limit reset if I upgrade?
Immediately. OpenAI's Free Tier FAQ states that if you reach a rate limit on a Free account, your usage rates reset when you upgrade to a Plus, Pro, or Business plan.
Can I make a custom GPT on the free plan?
No, and you can't on Go, Plus or Pro either; new GPT creation and publishing aren't available on any personal account. OpenAI is also retiring custom GPTs across ChatGPT plans, with a migration path to plugins. The standard retirement date is 11 December 2026, and existing GPTs remain usable until then.
Are custom GPTs going away?
Yes. OpenAI's 11 September 2026 release notes say it plans to retire custom GPTs across ChatGPT plans. The retirement FAQ sets 11 December 2026 as the standard date, and 11 February 2027 for FedRAMP workspaces and Enterprise workspaces with an approved deferral.
Comments