ChatGPT Pricing in 2026: Every Plan Change, With a Date and a Source

ChatGPT pricing in 2026 has exactly one fixed point. Plus cost $20 a month in January, and no report we opened shows it costing anything else since. Around that fixed point, OpenAI added three new price points, put ads on its two cheapest tiers, closed and reopened its $200 plan, and will halve that plan's included usage on 30 October.

The full 2026 ChatGPT pricing timeline, one row per change

The US consumer ladder, as of 10 October 2026, runs as follows. Free, then Go at $8, Plus at $20, and Pro at $100, $200 and $500 a month. Business is sold per seat. The biggest change of the year is not a price. It is the $200 Pro plan's Work and Codex allowance halving from 20 times Plus to 10 times Plus on 30 October.

That is the short answer. The longer one is the table below. Each row is a change to what a ChatGPT subscriber pays or receives, dated to the day where a source we opened gives one, and dated to a window where it does not.

Eleven rows in all. They hold four new price points (counting the Business Premium seat), one price cut whose size we could not verify, and one allowance cut announced a month before it takes effect. Rows dated to a window rather than a day are dated that way on purpose. The Weakest section below explains why.

ChatGPT Go and ads: the bottom of the range got cheaper and got sold

The year opened with an expansion, not a squeeze. On 16 January, Times of AI reported that ChatGPT Go was available worldwide, including the US, at $8 a month. Go first launched in India in August 2025, so 2026 was its arrival in the market that sets the reference price.

Go sits between Free and Plus. The same report described it as offering 10 times Free's messages, file uploads and image creation. That is a clean ladder: Free, Go at $8, Plus at $20, Pro at $200.

Ads arrived in the same announcement

The Go launch carried a second change. OpenAI said it would test ads for logged-in adults on Free and Go, and that paid plans from Plus upward would stay ad-free. By 9 April, TechCrunch was describing Free as a tier "which now includes ads" and Go as one "that also includes ads".

Axios headlined the start of ads on Free and Go on 9 February. We could not open that report, so the timeline dates the change to the window it must fall inside. That is less satisfying than a single date and more honest than borrowing one.

Read together, the two changes describe one product decision. Go is the cheapest ad-supported way to get more ChatGPT, and Plus is the cheapest way to make the ads go away. The $12 gap between them is partly a price for capacity and partly a price for an ad-free screen.

I think that second part is underrated. Ad-free is a feature now, and OpenAI can raise or lower the value of it without touching a single model. That gives the company a lever over Plus conversion that has nothing to do with compute cost.

April: the $100 ChatGPT Pro tier fills a hole in the price ladder

For more than a year, the jump from Plus to Pro was a jump from $20 to $200. On 9 April, OpenAI put a step in the middle. TechCrunch reported a new $100 Pro plan with 5 times the Codex usage of Plus, while the $200 plan kept 20 times Plus.

The framing was explicitly about coding. An OpenAI spokesperson told TechCrunch that more than 3 million people used Codex weekly, and that the tier was designed to give developers "more practical coding capacity for the money". The comparison drawn in the same statement was with Claude Code, not with any chat product.

Both Pro plans sold the same product

The detail that matters most is a quiet one. According to TechCrunch, both Pro plans offered the same core features, and "the main difference is the rate limits". That means the $100 step was not a cut-down product. It was the same product with a smaller meter.

That is the clearest statement of OpenAI's 2026 pricing logic anywhere in the record. At the top of the range, you are not buying features. You are buying a multiple of Plus usage. How Anthropic made the same move on Claude seats is covered in the breakdown of Anthropic's shift from seat price to usage.

TechCrunch also noticed something odd. OpenAI's pricing page at the time did not list a $200 plan at all, even though OpenAI confirmed the tier was still available. For anyone trying to rebuild this history from the official page, that is a warning. The page is a sales surface, not a ledger.

In early April, StreetInsider ran the headline "OpenAI introduces pay-as-you-go Codex seats, cuts ChatGPT Business price". We could not open the report, and every specific figure we found for the new seat price came from secondary pricing guides. So the cut appears in the timeline, and its size does not.

August: ChatGPT Plus limits and ads both moved, the price did not

By late August, Plus subscribers had been running on a weekly cap alone. On 24 August, 9to5Mac reported that Thibault Sottiaux, OpenAI's engineering lead for Codex and ChatGPT, would bring back the five-hour limit for Plus across ChatGPT Work and Codex the next day.

His stated reason was load management. The five-hour limit, he wrote, "allows us to smoothen the load on our compute, allowing to keep the plan generous in terms of weekly usage". Pro 100 and Pro 200 were kept exempt "for the upcoming months".

A five-hour window is a price change by another name

I would count this as a change to what Plus costs. The weekly allowance may be the same, but a subscriber who used to spend it in one long session now cannot. For a developer whose work arrives in bursts, a smoothed allowance is worth less than a lumpy one at the same $20.

This pattern is not unique to OpenAI. How allowance changes behave when the headline price stays put is the subject of the analysis of AI shrinkflation in usage limits.

Ads moved the same week. On 21 August, Euronews reported that ChatGPT ads would extend to 31 European countries from 24 August. The ads applied only to Free and Go. Plus, Pro and Enterprise stayed ad-free.

On 25 August, an announcement thread on the OpenAI Developer Community introduced a $100 Premium seat for ChatGPT Business. It is the Business equivalent of the April move: a higher-priced seat for heavier usage, sitting above the standard one.

September: the $200 ChatGPT Pro plan closes, reopens and halves

September is where the year's pricing story turns. On 10 September, OpenAI paused new sign-ups and upgrades to the $200 plan. CIO quoted OpenAI's help documentation: "as of September 10, 2026, we're temporarily pausing new sign-ups and upgrades to the ChatGPT Pro $200 plan (Pro 20X)." Existing subscribers were unaffected.

Sottiaux explained the pause on X. The $200 subscriptions, he wrote, "put the most strain on our systems", and "demand for Astra is really unprecedented". Astra is the GPT-6 model whose launch drove the demand. The $100 tier stayed open throughout.

DevDay: a $500 tier and a smaller $200 tier

Nineteen days later, at DevDay on 29 September, OpenAI reopened the $200 plan and changed what it contained. The Next Web reported a new Pro 500 plan at $500 a month, the only Pro tier with Ultrafast, which OpenAI said generates up to 300 tokens per second in Codex.

The same report set out the cut. From 30 October, Pro 200's included Work and Codex usage falls from 20 times to 10 times the Plus allowance. GPT-6 Pro messages in chat fall from 200 to 100 a week. Existing subscribers keep their limits until 29 October.

OpenAI paired the cut with two concessions. Existing Pro 200 subscribers receive a one-time grant of usage credits worth $2,500, which expires at the end of the year. OpenAI also said it would not reintroduce a five-hour limit on Pro 200, which removes the threat that hung over the plan after the August Plus change.

The credits are generous for a 61-day window. They are also a classic way to soften a permanent change with a temporary benefit. When the credits expire, the 10 times allowance is what remains.

ChatGPT Pro price per unit of usage converges on $20

Because OpenAI now prices its top tiers as multiples of Plus, you can compute a unit price. Divide the monthly price by the multiple of Plus usage it buys. The result is the cost of one Plus-sized allowance of Work and Codex usage at each tier.

Multiples are OpenAI's as reported by TechCrunch (April) and The Next Web (September). Secondary guides give a multiple for Pro 500, but we could not trace it to a source we opened, so it is excluded rather than estimated.

What disappearing volume discounts mean

Until 29 October, Pro 200 was the cheapest way to buy ChatGPT usage by a factor of 2. That is unusual in software, where the top tier normally carries the best unit rate. After 30 October, OpenAI's individual ladder becomes flat: you buy as many Plus-sized units as you need, at $20 each.

Flat unit pricing tells you what a vendor believes about its marginal cost. If serving the 20th unit costs about as much as serving the first, a volume discount is a subsidy to the heaviest users. OpenAI has stopped paying that subsidy. The broader trade between bundled and metered AI pricing is laid out in the comparison of bundling against metering.

What the 2026 ChatGPT pricing changes say about OpenAI's strategy

Taken one at a time, the eleven rows look like tactical adjustments. Taken together, they describe three clear choices. I would read each of them as deliberate.

Codex, not chat, drives the top of the range

Every top-tier change this year was framed around coding. The April tier was sold as Codex capacity, the August limit applied to Work and Codex, and Pro 500's distinguishing feature is Codex speed. Chat is now the cheap end of the business. Agentic coding is the part OpenAI prices with care.

That matters because coding agents burn far more compute per user than chat. A plan priced for chat-era usage cannot survive agent-era usage at the same multiple, which is the cleanest explanation for the Pro 200 cut. How compute-hungry products erode gross margins is covered in the analysis of real AI gross margins.

It is tempting to read every change as OpenAI squeezing subscribers ahead of a listing. I do not think the record supports that on its own. The bottom of the range got cheaper this year, and a $100 option appeared where none had existed. The squeeze is concentrated on one plan whose unit price had been half everyone else's. The question of how OpenAI's revenue multiple shapes these choices sits in the breakdown of OpenAI's valuation against its run rate.

Where this record of ChatGPT pricing is weakest

A pricing history is only as good as its sources. This one has three specific weaknesses, stated here rather than buried.

We did not read the archived pricing pages directly

The goal for this post was to rebuild the year from archived snapshots of OpenAI's pricing page. During research, OpenAI's pricing and help pages returned access errors to our tools, and the main web archive could not be reached. So every row rests on dated reporting that quotes OpenAI's statements, help documentation or staff, not on the archived page itself. That is a real gap, and the next revision of this post should close it.

The April episode shows why the gap matters less than it sounds. The official page omitted a plan that was on sale. An archive of that page would have recorded the omission faithfully, and been wrong about what you could buy.

Some rows are thinner than others

The Business price cut appears without a size. The ads launch appears as a window rather than a day. The Business Premium seat rests on a community announcement thread rather than a press release. Pro 500's usage multiple is left out entirely. Each of those is a choice to publish a visible gap rather than a borrowed number.

We also found one widely indexed page claiming Plus rose to $24 in May. No other source we found supports it, and every dated report after May still treats Plus as a $20 plan. It is not in the timeline. It is a useful reminder that a pricing history scraped from aggregators inherits their errors.

"Plus usage" is a moving unit

Every multiple in this post is a multiple of Plus, and Plus itself changed in August. A 10 times allowance is 10 times a number OpenAI can adjust. The unit-price table is therefore a statement about OpenAI's internal ratios, not a fixed quantity of compute. Treat it as a comparison between tiers at a moment, not a constant. The case for publishing rate limits as hard numbers is set out in the proposal for a rate-limit transparency standard.

Frequently asked questions

Where to start if your team pays for ChatGPT

If anyone on your team holds a Pro 200 plan, put a date in the calendar for 22 October. Pull that person's last four weeks of Work and Codex usage and compare it against half of their current allowance. If they routinely use more than half, the 30 October change will hit them. The choice between Pro 500, a second account or a slower workflow is better made before the cut than during it.

Then start keeping your own record. Each time you renew or change a plan, save the pricing page and the relevant help article with the date. This year showed that the vendor's page will not do it for you. How to argue renewal terms once you hold that record is covered in the guide to grandfathered pricing at renewal.

Related reading