If you pay for ChatGPT, or you've been thinking about it, the last month has been confusing. In the space of three weeks OpenAI stopped selling its $200 plan, brought it back with half the usage, and added a new plan at $500 a month. The price of the $200 plan never moved. What you get for it did.

That gap between the price you see and what's inside it is the real story here, and it's worth understanding properly. Not because most people need a $500 AI subscription (they don't), but because every AI company now prices this way. Once you can read one plan, you can read all of them, and you stop paying for usage you never touch.

What happened to the 20x plan

On 10 September 2026, OpenAI stopped taking new sign-ups and upgrades for ChatGPT Pro at $200 a month, the plan known as Pro 20x because it gave you roughly 20 times the usage of ChatGPT Plus. If you were already on it, nothing changed. The $100 Pro plan, which gives five times Plus, stayed on sale. But if you cancelled or downgraded from the $200 plan, you couldn't buy it back while the pause lasted.

The reason was capacity. OpenAI's newest model, GPT-6 Astra, was in heavy demand, and users across social media were reporting that the model they'd selected was "at capacity". The $200 plan was the obvious pressure point. It had the fewest subscribers, but those subscribers were the ones most likely to use every message they'd paid for. It was also the best value on the menu: twice the price of the $100 plan for four times the usage.

That last point matters. When a company stops selling its best-value plan, it is usually telling you that plan was costing it more than it earned.

How it came back, with new tiers

At its DevDay event on 29 September, OpenAI reopened the $200 plan and added a third rung above it. Pro now comes in three monthly plans, named after their price, all measured against the $20 Plus plan:

• Plus, $20: the baseline everything else is measured against.

• Pro 100, $100: five times Plus. The entry Pro plan, with Pro models, Codex and deep research.

• Pro 200, $200: ten times Plus, down from twenty, with GPT-6 Pro messages cut from 200 to 100 a week.

• Pro 500, $500: 25 times Plus, and the only Pro plan with Ultrafast, a speed mode in ChatGPT Work and Codex.

The $200 plan is the one that changed. New subscribers now get 10 times Plus, not 20. OpenAI's help centre says the lower allowance reflects its increasingly efficient models, which is one way of putting it.

If you already had the $200 plan, you got a short reprieve. Anyone with an active Pro 200 subscription between 22 September and the morning of 29 September keeps the old allowance until 29 October. From 30 October, everyone on Pro 200 is on the same 10x plan at the same $200. Affected subscribers were also given a one-off batch of usage credits that expire on 31 December 2026.

Notice what the new menu does. Before, the $200 plan worked out at about $10 for each "Plus's worth" of usage. Now every Pro plan costs the same $20 per unit: $100 for 5x, $200 for 10x, $500 for 25x. The discount for buying more has gone. What you pay extra for at the top is speed.

How to read an AI price plan

This is the part worth keeping, because it applies to ChatGPT, Claude, Gemini and whatever comes next. Five questions will tell you more than the headline price.

1. What is the multiplier measured against? "10x" is not an amount. It means ten times the Plus allowance, and the Plus allowance itself can move. When one company says 20x and another says 10x, they may be measuring against different baselines and different models, so you can't compare them directly.

2. What does one unit cost? Divide the price by the multiplier. It's the quickest way to see where the value sits on a menu, and the quickest way to spot when that value quietly changes. Pro 200 went from $10 a unit to $20 without its price moving.

3. Do some features use your allowance faster? On ChatGPT Pro, Fast mode uses your allowance at 2.5 times the normal rate, and Ultrafast uses it at eight times. A faster answer is also a more expensive one, even when it doesn't show on your bill.

4. What do top-up credits actually cover, and when do they expire? OpenAI's credits work in Codex, ChatGPT Work and its Office add-ins, but not for every model or feature. Credits handed out as compensation have an end date. Unused, they're worth nothing.

5. What are the dates? Grandfathering means keeping old terms for a while. It always has a cut-off, and the cut-off is the date your plan really changes, not the day of the announcement.

What's happening to AI pricing

Step back from ChatGPT and the pattern is the same everywhere. The entry price has settled at about $20 a month: ChatGPT Plus, Claude Pro and Google AI Pro all sit there. Above it, each company now runs a ladder. Anthropic sells Claude Max at $100 for five times Pro and $200 for twenty times. In May, Google cut its top plan, AI Ultra, from $249.99 to $100, and moved from daily prompt limits to a compute-based allowance that refreshes every five hours. Below the $20 line, ChatGPT Go costs about $8 in the US, and OpenAI has been testing ads on its Free and Go tiers.

So prices aren't simply going up or down. The sticker prices are stable, and the thing inside them is being measured more carefully. Anthropic has said openly that computing capacity, not price, has been the real constraint this year. OpenAI's pause in September was the same problem made visible.

What happens next is a reading of the signs, not a forecast anyone has confirmed. Three things look likely. First, more plans will measure the work rather than count the messages, so a quick question costs less than a long research task or a video. Second, speed is becoming something you pay for on its own, as Pro 500's Ultrafast shows. Third, the cheapest tiers will lean more on ads and the top tiers will lean on top-up credits. For most people, the $20 plan will stay the sensible choice. The risk is paying for a big plan out of habit, or because one bad afternoon of hitting a limit made you panic.

Work out which plan you actually need

You don't need to guess. Give it one ordinary working week and keep a note on your phone.

• Each time you hit a limit, write down the day, the time and what you were doing.

• Note whether you were stuck, or whether waiting an hour or switching model would have been fine.

• At the end of the week, count the limits that genuinely stopped your work.

If the answer is none, or once or twice, stay on the $20 plan (or try the cheaper tier if you only use AI lightly), unless you need integrations that only come with a higher plan. If connecting AI to the other tools you work in is the point, you're paying for access rather than usage, so check the integration you need is actually included before you upgrade.

If you hit a wall most days on work that earns you money, one step up is justified, and five times the usage is usually plenty. The top tiers are built for people running coding agents for hours at a time. If that isn't you, the extra mostly goes unused.

The bigger saving is in how you use whichever plan you're on. Long, rambling chats that repeat your context every time burn through an allowance far faster than a setup that already knows who you are, what you work on and how you like things done. That's how I set things up in AI for Creators: your AI learns your work once, so every message goes further. Plans will keep changing. A good setup moves with you from one to the next.

Prices are US list prices as of 11 October 2026. UK prices differ.