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. If your free account behaved strangely in early August, you weren't imagining it; OpenAI swapped the default model underneath the Free tier mid-week, and for five days its own documentation contradicted itself about what you were getting. This page watched that happen, and the record of it is below.

Last verified: 8 August 2026; rollout status, Free/Go limits and storage figures re-verified 11 August 2026. Every figure below was checked against an OpenAI help page or an OpenAI blog post on one of those dates. 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 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 says the same for Go, and documents Think as live: "Free and Go users can use Think for harder questions in ChatGPT on the web and in the mobile app." One oddity worth noting: the announcement post itself was never updated — it still promises in future tense what the help pages now document 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, and the announcement said plainly that "Limits will still apply for file uploads, images and other tools."

One piece is still officially mid-flight: the Luna default. The Availability section states it as fact — "Their default model is GPT-5.6 Luna" — while the Overview on the same page still says Luna "is becoming" the default, and a later section adds that Luna "powers the default ChatGPT experience for Free and Go users as it rolls out". Everything else in this section is present tense on OpenAI's own pages as of 11 August 2026.

The verified price and limit ladder

Two things in that table are worth more than the rest. The first is that you cannot buy a discount by committing to a year on any individual plan — OpenAI's Pro tiers page states flatly that "Currently, we do not support annual billing or the option to pay for multiple months in advance for ChatGPT Go, Plus, or Pro subscriptions." All three, not just Pro. Business is the only plan here with an annual rate at all, and it's the one aimed at teams. The second is that both Pro tiers "include the same core capabilities. The main difference is usage allowance: Pro $100 unlocks 5x higher usage than Plus, while Pro $200 unlocks 20x usage than Plus." Note OpenAI's word is main, not only — the same page adds that "allowances can differ by Pro tier," and the voice row above is a live example, since unlimited GPT-Live-1 is a $200 thing. Mostly, though, the extra hundred dollars buys headroom rather than capability.

Web search stays off that table because OpenAI only publishes it as a Free-tier capability, and it's a good one: "Search the web and obtain up-to-date information" is listed outright on the Free Tier FAQ. You don't have to pay to get current information out of ChatGPT.

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 single widest gap between any two rungs of this ladder. Projects allow 5 files per project on Free and 40 on Pro and Business. The 10-file simultaneous upload ceiling applies to everyone including the $200 tier, but note that it's a Projects limit specifically, not a global one.

One caveat on the Plus row, in keeping with the rest of this page: OpenAI publishes two different numbers for it. The Projects help article says 25 files per project; the File Uploads FAQ says "up to 20 files per project" for Plus. Both pages were updated within the last fortnight. I don't know which is current, so the table carries both rather than picking the flattering one.

For five days, OpenAI's own help page contradicted itself — here's the record

This section used to be written in the present tense. It's worth keeping in the past tense, because it explains what free users lived through in early August, and because watching a vendor's documentation disagree with itself is the clearest picture you'll ever get of how these rollouts actually work.

The article at help.openai.com/en/articles/20001354-gpt-56-in-chatgpt says two incompatible things about the Free tier, in two different sections of the same page.

The Overview opens with: "GPT-5.6 Luna is becoming the default model for Free and Go users."

The Availability by ChatGPT plan section agrees: "Free and Go users do not have access to GPT-5.6 Sol. Their default model is GPT-5.6 Luna, which also powers Think." But the line introducing that same section's table — a few lines above it, in the same section — reads: "GPT-5.5 Instant is not shown because it remains the default model."

And the Instant limits section further down still describes Free access the old way: "Free-tier users have limited GPT-5.5 Instant access within a 5-hour window."

So on 8 August the page stated that Luna was becoming the Free default and that GPT-5.5 Instant remained the default model, within a few lines of each other, and then metered the Free tier against Instant in a third place. All three passages were live simultaneously.

By 11 August, OpenAI had resolved it. The page (stamped "Updated: 2 hours ago" when I re-checked) deleted both stale legs — "GPT-5.5 Instant is not shown because it remains the default model" and the "limited GPT-5.5 Instant access within a 5-hour window" sentence are gone — and the Instant-limits section now reads: "Free and Go users have unlimited everyday text chats, subject to abuse-prevention safeguards." What remains is only a tense mix (the Overview still says Luna "is becoming" the default, and the Work/Codex/API section says it powers the experience "as it rolls out", while Availability states flatly that it is the default), plus one orphaned sentence: the Free Tier FAQ still warns that "GPT usage follows your current Free tier model limit" without defining what that limit is now that text chats are unlimited. If your free account hit a wall in early August, this five-day window is almost certainly why — the change was in flight, not your account broken.

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.

I could not find either figure on any of OpenAI's current limits pages — the Free Tier FAQ, the GPT-5.6 page, the Plus page or the File Uploads FAQ, all checked again on 11 August 2026. (A 2024 release-note archive does mention two DALL·E 3 images a day for Free, but that model retires 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: "The following tools can have usage limits separate from the main chat limit: Data analysis with ChatGPT, File and image uploads, Create images in ChatGPT." And elsewhere on the same page: "GPT usage on the Free tier follows the same plan limits as ChatGPT. Advanced functionality, including data analysis, file uploads, and image creation, can have stricter limits than paid plans."

"Can have usage limits" is a deliberate non-answer — and since the August rollout, the main-chat side of the question has dissolved entirely: there is no text-chat cap left to publish.

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, 300 messages every three hours for Go users with Think switched on (a figure that changed with the August rollout — it was 10 per 5 hours days earlier), 3,000 GPT-5.5 Thinking requests a week on Business. Those are specific, and they're published.

So the silence is narrow and it's deliberate. The two figures OpenAI declines to state are exactly the two the whole internet quotes: how many messages a free user gets per 5 hours, and how many images per day. Those move with load, which is the obvious reason not to 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. We took the same approach when we checked whether the Claude API is free: several vendors have quietly stopped publishing free-tier numbers, and the correct response is to say so.

A few more things I deliberately left out of the table because I couldn't verify them:

Deep Research on Free. OpenAI's FAQ confirms allowances "vary by plan" and points you at an in-product counter. It doesn't state a Free-tier number, so I don't have one. 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.

Connectors and memory on Free. No primary page I could fetch states the Free-tier scope for either. Unconfirmed, so unstated.

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 is being retired on 30 August 2026 — the release note advises downloading anything you want to keep before then. That's three weeks from this update, so if you have image history sitting in that GPT, go get it.

The context window caveat

GPT-5.6 Luna's documented context window is 128K, with GPT-5.6 Terra also at 128K and GPT-5.6 Sol at 272K. Important qualifier: that figure is published on the ChatGPT Business plan's models-and-limits page, not on any Free-tier page. Luna is the Free/Go default, so 128K is the model's window — but OpenAI does not state anywhere I could find that the Free plan is entitled to the full 128K in practice. Treat it as a model-level number, not a free-plan guarantee.

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 unusually clean for once. Go at $8 buys you the GPT creator, 25 files per project instead of 5, 4 GB of Library storage instead of 500 MB, and real voice time. Plus at $20 buys the same voice allowance as Go and a Projects cap that OpenAI states as either the same or slightly lower — so the honest case for Plus over Go is model access, usage headroom, and 20 GB of storage against Go's 4 GB. Pro's two rungs share the same core capabilities and differ mainly in allowance, with 100 GB of storage and unlimited live voice at $200. 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. Two caveats if you're buying seats. Business has a two-seat minimum, so $20 a head means $40 a month, not $20. And a second seat type is on the way: OpenAI's Business page now announces that "Premium seats cost $100 per user per month when billed annually, or $125 per user per month when billed monthly," including "5x more usage than Standard seats, no 5-hour usage limit, and the flexibility to mix, assign, and reassign seat types." It has not launched — there is a waitlist, open "before August 20" — so the $25/$20 figures above are Standard-seat prices and remain current. Note the detail hiding in that announcement. Premium promising "no 5-hour usage limit" only means something if Standard seats have one — and OpenAI's Premium seats announcement says it twice more, that Premium "removes the five-hour usage limit" and, in its feature list, "No five-hour usage limit." So across two OpenAI pages the five-hour window is confirmed to exist for Standard seats, and on neither one is it given a number. That is the single clearest example on this page of the pattern I keep running into: OpenAI will confirm a limit exists while declining to say what it is.

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; our free AI APIs roundup covers the metered side, and the tools I actually use daily piece covers what earns a subscription.

The rollout landed, the contradiction resolved, and this page was re-verified against the updated pages on 11 August 2026. The next scheduled re-check is the monthly refresh — sooner if the orphaned "Free tier model limit" sentence gets a definition.