How Much Does ChatGPT Cost — Plus / Pro / Team / API Pricing Compared

ChatGPT subscriptions come in five tiers: Free at $0, Plus at $20/month, Pro at $200/month, Team from $25/seat/month, and API billed separately by usage. For an individual asking how much ChatGPT costs per month, the standard answer is $20. But that is only the list price — the number that leaves your card is not necessarily a clean $20: web and Android are not even the same price, and the state on your billing address may add sales tax on top. This article lays out every paid OpenAI tier in one pass and uses the charge amounts that actually show up in our own ledger to answer the question the pricing page will not —how much am I really going to be charged.

Prices verified as of September 2026. OpenAI adjusts pricing and regional tiers, so the amount shown on the Upgrade page inside your own account is authoritative. Every "real charge amount" and "failure-reason share" in this article comes from our own ledger (through 2026-07-17), the same source as theChatGPT payment compatibility page on this site. It gives price points and percentages only, and does not claim to represent the whole market.

1. The Full ChatGPT Price Table (Every Tier in One Place)

Conclusion first: for an individual, Plus ($20/month) is enough. The other four tiers only make sense in specific situations. This comparison table puts "list price" and "what we actually see charged in our ledger" side by side, because those two columns are frequently different — and why they differ is the subject of section 2.

On sample size: the Plus tier has ample sample in our ledger and a stable price distribution; Team is moderate; Pro is thin, good only as evidence that the price point exists, not enough to discuss a distribution.

2. How Much ChatGPT Costs per Month: What You Will Actually Be Charged

Conclusion: the standard web charges are $20.00 and $21.28 (the latter including state tax), Team lands at $25.00/seat, and Android's sticker price is $19.99. Two different kinds of numbers have to be kept apart here: $20.00 / $21.28 / $24-25 come from our own ledger (reconciled transaction by transaction, through 2026-07-17) and are amounts really taken off a card; $19.99 is a Google Play price-tier value, a publicly listed external price, not something observed in our ledger. The spread between them is not somebody overcharging you — it is that you are not on the same payment rail, and that the state on your billing address decides whether tax gets added.

One more price point deserves to be called out on its own: $24-25. It is not a price increase and it is not tax —it is an arrears catch-up: when an invoice on the account went unsettled, the next charge collects the old balance along with the new one. This shape shows up repeatedly in the ledger and is the most common "not enough funded" trap: you funded for $20 or $21, a $24-25 charge lands, and it is declined outright. The funding advice in section 7 is derived backwards from this most expensive shape.

Why web is $20.00 and Android is $19.99

Because they are two entirely separate payment rails:

• Web (chatgpt.com → Upgrade): the charge is initiated through Stripe and the merchant is OpenAI itself. The price is $20.00, with tax calculated by OpenAI from your billing address and added on top.
• Android (in-app purchase): the charge is initiated by Google Play and the merchant is Google. Pricing follows Google Play's price tiers, landing on the standard ".99" price point of $19.99 (that is Google Play's list price, not a charge shape from our ledger). Tax is handled by Google based on the region it has on file.

A one-cent difference means nothing on its own. What does matter is that the failure reasons change with the rail: a web decline gives you reason codes from Stripe and the issuer; an Android decline shows you Google Play's "payment method was declined", and Google's validation rules for a stored card — especially country/region consistency — are not Stripe's. The same card charging successfully on web while refusing to attach in the Play Store is entirely possible — and in that case swapping cards usually will not help, switching rails will. For the full decline-reason walkthrough, see the top 10 reasons a virtual card gets declined.

Why some people are charged more than $20

Several US states levy sales tax on digital services, and OpenAI calculates whether tax applies, and how much, from the state on your billing address. So the same Plus subscription lands differently for different people: some pay a clean $20.00, others pay $21.28 with tax — that is not an overcharge, it is tax. The gap is set by tax law, not a variable you can work; fold it into what you fund and move on.

How to check what you were actually charged

1. On the OpenAI side: Settings → My plan → Manage my subscription takes you into the Stripe billing portal, where every period's invoice amount and tax breakdown is visible.
2. On the card side: open the transaction list on the card detail page and look at the actual amount of that authorization. The two rarely disagree; when they do, first check whether you are reading an authorization and a settlement as two separate charges.
3. Timing gap: a subscription authorizes first and settles later, so the available limit on the card is held first and the final settled amount may differ slightly from the authorized amount (usually a tax delta).

3. What $20 a Month for ChatGPT Plus Actually Buys

Conclusion: Plus buys the newest models, higher usage ceilings and the full tool set — it does not buy unlimited use. This matters for the price judgment: many people assume $20 means "use it however you want", when Plus also has usage windows. They are just far higher than the free tier's.

What Plus actually changes versus the free tier:

• Models: access to the strongest current reasoning model, where the free tier usually gets the previous generation or a downgraded variant.
• Usage ceilings: the free tier hits a wall after a few messages; the Plus window is much larger, but it still exists.
• Tools: image generation, data analysis / code execution, web search, voice, file upload, custom GPTs — practically all of it unlocks at the Plus tier.
• Priority at peak: Plus users go first when the service is busy.

Which makes "is $20 worth it" easy to answer: have you ever hit the free tier's usage wall? If you do not hit it once a week, free is fine; if you hit it twice a day, $20 is probably the best value subscription you will buy this year. For the subscription steps themselves (sign-up, region, card, upgrade) see the complete ChatGPT Plus subscription guide; this article does not repeat them.

4. ChatGPT Pro at $200 a Month: Who Actually Needs It

Conclusion: Pro costs 10x Plus, but what it solves is not "more features" — it is "never being interrupted by usage limits" again. If you have never had to stop mid-task because your allowance ran out, Pro is meaningless to you.

Pro's value concentrates in three things:

• Heavy use of the strongest reasoning mode — Plus restricts that most expensive mode the hardest, Pro opens it up the most.
• Long continuous sessions without interruption — people writing long pieces, running large batch analyses or debugging code for hours feel this most.
• Higher priority and earlier access to new features.

A practical way to decide: convert $200 into your own hourly rate. If the time you lose each month to waiting, switching tools and redoing work because of usage limits exceeds the hours $200 buys, Pro pays for itself; otherwise Plus is the reasonable choice, or go with API usage billing from section 6.

5. ChatGPT Team at $25/User/Month: Is Splitting a Plan Really Cheaper

Conclusion: Team costs more per person than Plus ($25 vs $20). What it makes cheaper is management overhead, not price. Treating Team as a way to split costs is the wrong frame.

Plenty of guides sell Team as a "shared-plan hack" on the grounds that "several people share one workspace". The billing model has to be stated plainly:

• Team bills per seat, not per workspace. Pull 5 people in and that is 5 seats, multiplied head by head on the invoice. The Team charge shape in our ledger is exactly the $25.00 tier, counted per user.
• There is a seat minimum. Team starts at 2 seats — one person cannot open a Team workspace.
• Annual and monthly are two different unit prices. $25/seat/month is the annual-plan unit price; the monthly unit price is higher. This is one of the few places in OpenAI's lineup with a genuine annual discount, at the cost of paying a year up front. Every calculation below uses the lower $25 annual-equivalent price — the conclusion only gets worse on monthly billing, and the actual quote on your Upgrade page is authoritative.
• What Team buys is workspace capability: shared custom GPTs, member management, and the policy of not training on your data by default. That is worth something to a company. It is worth nothing to "three friends splitting one account".

If you still want to split it, settle these three things first

1. The workspace belongs to whoever created it. If the person paying and the person owning the workspace are not the same, agree up front who keeps the assets and history when you part ways.
2. Several people logging in from several places is a risk signal. When several people log in to one account from very different network environments and switch between them often, that is a textbook anomaly pattern. For how the penalties play out, see the real reasons Claude / ChatGPT accounts get banned.
3. Run the numbers before deciding. At the $25 annual-equivalent unit price, two people on Team is $50/month; two people each on their own Plus is $40/month. And that is Team at its cheapest — on monthly billing the per-seat price is higher and the gap only widens. Unless you genuinely need the workspace features, buying separately costs less.

6. How the OpenAI API Is Billed: A Completely Separate Account From Subscriptions

Conclusion: the API is prepaid credit drawn down by token, and it does not connect to a ChatGPT subscription in either direction — your Plus fee grants you no API credit, and money you deposit into the API cannot be applied to Plus. This is the most common and most expensive misunderstanding for newcomers: people subscribe to Plus "to use the API" and then find their API balance is still $0.

The API runs on the separate platform.openai.com back end, with its own card on file and its own top-up. The API-side charges we see in our ledger cluster on a few round top-up tiers: $5 / $10 / $100 / $200 — which are the common presets on the top-up page, rather than "a little deducted per call". Put another way,with the API you put money in first, then burn it down by token.

For very light usage the API can beat $20/month: a few dozen questions a month and $5 lasts a long while. The trade-off is that you have to run a client and manage your own balance, and you lose the out-of-the-box Plus tooling (images, code execution, web search). For heavy users the API is usually far more expensive than a subscription — a subscription is a fixed price, and tokens have no ceiling.

7. Hidden Costs: What a $20 Plus Subscription Really Requires You to Fund

Conclusion: $20 is what the merchant collects, not what you need to have ready. If you are in China paying with a virtual card, the real cost also includes the account top-up fee, the card issuance fee, and the headroom you have to leave on the card. This section breaks out every line — the part almost every "ChatGPT pricing" article leaves out entirely.

Three details that genuinely move your total

1. The account top-up fee is proportional, so splitting top-ups changes nothing. The 0.5% has no tiers and no per-transaction threshold: one top-up of $420 and 12 top-ups of $35 (the per-transaction minimum, per the top-up page) both cost $2.10 in fees.What does cost you more is opening cards over and over — that fee is one-off and charged per card. And "a new card every month" is not a smooth path anyway:by default you can hold at most 5 active cards, with a cumulative cap of 5 as well (closed cards included); once you are at the cap you can request an increase on the card-issuing page, and whether it is approved is decided automatically from your account's usage history, and the number of cards you have left is whatever the card-issuing page shows. For a long-running subscription, the least painful approach is still one card all the way through — a new card every month means re-attaching a card every month, and attaching a card is exactly where failures cluster.
2. A card's limit is a one-time pool set at issuance, and renewals keep eating into it. Plenty of people subscribe successfully once and get declined in month two right here: the limit was set for "$20 is enough", and by the second charge the pool was empty. If you plan to subscribe for a year, plan the total limit as 12 × $26 ≈ $310 (covering tax and the arrears catch-up shape together), then leave a little room for per-transaction fees, rather than sizing it to a single month's $20. If the limit runs short you can top the card up at any time; rates are in the table above.
3. To pull the leftover money off the card after canceling, you must wait 48 hours from the last transaction. Pulling the balance back to your account costs nothing, but this 48-hour window gets overlooked constantly, producing the "I clicked pull-back and nothing happened" confusion. Some BIN ranges also require a small balance to stay on the card; the exact amount is whatever the card detail page shows.

So how much do you actually need to fund for one Plus subscription

Run the lines above along the minimum viable path and the numbers close:

• Top up the account by 35 USD (the current per-transaction minimum, per the top-up page); after the 0.5% fee, $34.82 is credited, usually within minutes.
• Open a card with a 30 limit (30 USD is the current minimum issuance limit, per the card-issuing page); issuance takes seconds. Three amounts come off the account: the limit itself, $30, plus the issuance fee (from $1 per card, varying by BIN range; the shared-limit card recommended for AI subscriptions is $2, shown on the card-issuing page) plus the limit service fee of $0.60 (at the 2% tier) = at most $32.60. Note that the limit reaches the card in full — available on the card is $30.00, and the service fee does not come out of the limit.
• That leaves roughly $2.22 in your account balance, usable directly the next time you top the card up.

Target on the card: above $26 (covering the $24–25 arrears catch-up shape specific to ChatGPT). That number is not arbitrary — it has to survive all three real charge shapes of the $20 plan at once: a clean $20.00, $21.28 with state tax, and the $24-25 arrears catch-up, with a little room left for per-transaction fees. And since the minimum issuance limit is 30 USD anyway, simply opening a minimum-limit card means month one takes care of itself; what needs attention is month two onward — the limit is a one-time pool, see point 2 above. If you are subscribing to the $200 tier, the same logic gives you $210 or more on the card.

Why fuss over a few dollars: failures split into two layers

Subscription charge failures fall into two layers, and the order cannot be reversed. Most people start by checking limit, balance and network one by one and skip the layer that sits in front of all of it.

If both layers check out and you are still declined, work through the reason codes one by one; the method is in troubleshooting a failed ChatGPT subscription paid from a Chinese card (that one is organized by the literal error text, and applies just as well to Claude and other AI services).

8. How to Buy ChatGPT for Less: 5 Popular Routes, Fact-Checked

Conclusion: only two actually save money (dropping to API usage billing, and taking Team annual when a team genuinely needs it). The other three either do not exist or trade a much larger risk for a few dollars. One at a time.

About "paid-on-your-behalf" services and cheap shared seats

Search results always turn up "we'll pay for you", "private account", "shared seat" offers below the official price. Only the mechanics here, no names: such services either hand you access to somebody else's account (you can be kicked off at any time, and the conversation data is not yours), or they pay on your behalf (you pay in RMB, they pay in dollars, and they take the spread and the exchange rate).The risk in the first is your account; the risk in the second is whether the other party stays in business. What both share is that you end up with no payment method of your own — next price rise, next service (Claude, Cursor, Midjourney and so on), you go looking for somebody again.

Which is why this article does not treat "saving those few dollars" as the goal. In the real cost structure of subscribing to AI tools, the monthly fee is the small part; whether you can pay reliably is the big part — the cost of one failed charge, in service interruption, re-subscribing, even account risk flags, far exceeds $20. Getting a card of your own on a BIN range that supports AI subscription scenarios pays off not this month, but on every renewal after it.

9. Pick by User Type: Which Tier Should You Actually Buy

Conclusion up front: students and light users should stay free until they hit the wall, then move to Plus; office workers and creators should go straight to Plus; developers should start with the API; only people who make a living on it and hit limits daily should consider Pro; companies that need data separation take Team.

If you are also looking at Claude, the price structure is completely different (Claude has three tiers — Pro, Max 5x, Max 20x — and its allowance math is not the same as ChatGPT's). For that comparison see how to choose a Claude subscription and how Claude usage limits are calculated. In short: for people who hit limits on both, the upgrade path differs — with ChatGPT you jump up to the $200 Pro tier, with Claude you pick a multiplier between the two Max tiers.

10. Paying From China: The 6-Step Flow

Conclusion: Chinese bank cards (including dual-currency cards carrying a Visa / Mastercard logo, and the various prepaid or multi-currency cards issued domestically) essentially do not work at OpenAI. The path that actually works is all three of: a US-BIN virtual card, its matching billing address, and a US network environment. Here is the full sequence for getting the money out the door.

1. Step 1: decide which tier you are subscribing to

   Use the table in section 9. This decides how much money you need next:fund the card to $26 for Plus (covering the $24–25 arrears catch-up shape specific to ChatGPT; the 30 USD minimum issuance limit already covers it), $210 for Pro, and per-seat price × seat count for Team. On the account side the minimum is 35 USD per top-up (per the top-up page).

2. Step 2: pick a BIN range that supports AI subscription scenarios and open a US-BIN virtual card

   The first thing, always, is to check which scenarios the BIN range supports — the issuer's policy decides whether that range clears AI subscription merchants, and the BIN range notes on the card-issuing page spell it out. Pick the wrong range and every subsequent diagnostic step is wasted effort. The second thing is to size the limit against total spend — the limit is a one-time pool, not a monthly reset, so if you plan to subscribe for a year, plan it by the year, not by the month. Once the balance is in place, the card is issued in seconds, with no waiting. Also: by default you can hold at most 5 active cards, with a cumulative cap of 5 (closed cards included); at the cap you can request an increase on the card-issuing page, with approval decided automatically from your account's usage history, and the number of cards you have left is whatever the card-issuing page shows.

3. Step 3: fund it properly and leave headroom

   The account top-up fee is 0.5% (no tiers; 35 USD minimum per top-up with 34.82 credited, usually within minutes; the minimum is per the top-up page);the card top-up service fee starts at 2% and drops to 0.5% on calendar-month cumulative card top-up volume, independent of BIN range. Per section 7, keep more than $26 on the card (covering $20.00 and $21.28, plus the $24–25 arrears catch-up shape specific to ChatGPT), and leave room for per-transaction fees, with the rates on the card-issuing page's fee details being authoritative.Do not fund exactly $20 — being short by a dollar and being short by a hundred produce the same decline.

4. Step 4: align four things

   Your account's region, the card's BIN country, the billing address, and the network exit you pay from — all four must be US and consistent with each other, and specifically the billing address and the IP you pay from must at least share a country, ideally a state. For the billing address, copy the set shown on the card detail page character for character into the checkout page: street only on line 1, line 2 empty, city / state / ZIP internally consistent. ⛔ Do not go find an address online to enter — a "universal address" shared by thousands of people is a textbook risk signal. As for whether you may change that set yourself and whether changing it affects payment, that varies by BIN range, and the card detail page and its notes are authoritative.

5. Step 5: subscribe on the web, not inside the app

   Use the Upgrade entry on chatgpt.com (the Stripe rail) — in our data this is the mainstream path and the one with the most readable reason codes. Use the copy buttons on the card detail page for the card number, expiry and CVV, and do not transcribe by hand — one wrong digit produces exactly the same outcome as no money: a decline.

6. Step 6: verify the amount after the charge

   Once it succeeds, go back to the transaction list on the card detail page and confirm whether the charge was $20.00, $21.28 (with tax) or $24-25 (with an arrears catch-up). If it fails, read the reason in the transaction detail before doing anything, and do not hammer retry:fix the cause, wait at least an hour, and never make a third consecutive attempt. Repeated failures in a short window get flagged as anomalous on the merchant side, and the more you click the harder it gets; on top of that, a dedicated-limit card is charged another $0.60 per failure (shared-limit cards are exempt), so hammering retry is literally paying to hit a wall. We have seen a real case on this site of "declined 8 times in 7 minutes on the same cause, then one success 5 minutes after stopping and raising the limit" (see three real AI subscription payment failures) — if the cause is not fixed, a hundred retries give the same result.
   On the "three attempts in a row triggers a lock" rule of thumb: the payment gateway's actual thresholds and cooldown durations are external parameters and are never published, so this is a value observed over time on our support side, not an official figure. Precisely because it is unverifiable, it is worth following on the conservative side.

The full sign-up, region-switching and card-attaching detail lives in the complete ChatGPT Plus subscription guide; if you are stuck at phone verification during sign-up, see the truth about ChatGPT phone verification.

11. FAQ

12. Related Reading

• ▸ The complete ChatGPT Plus subscription guide — the next step after pricing: sign-up, region, card, upgrade, end to end
• ▸ Troubleshooting a failed ChatGPT subscription paid from a Chinese card — working through reason codes when a charge is declined
• ▸ What is a virtual credit card — start here if you have never used one
• ▸ How to choose a Claude subscription (Pro / Max 5x / Max 20x) — the other vendor's price structure, compared
• ▸ Getting Claude Code running from China — subscription and payment for the developer case