
ChatGPT Plus costs $20 a month, billed monthly, and OpenAI sells no yearly version of it. Below it sits Go at $8 a month in the US; above it, Pro from $100. Whether the $20 pays off comes down to one thing: how much of Plus’s usage meter your week burns. Plain chat barely touches it, while reasoning models, Codex and agent work do.
We read that meter on our own Plus seat for four months, 293 readings, while it ran coding agents through Codex. One full five-hour window took about 15% of the week. Priced at OpenAI’s own token rates, a whole Plus week held about $72 of work. That makes a month worth around $310 for your $20, if you use all of it.
ChatGPT Plus price and plans in November 2026
| Plan | Price | What it adds |
|---|---|---|
| Free | $0 | Unlimited text chats with GPT-6; limited uploads, images, voice and deep research; Codex only on GPT-6 Luna |
| Go | $8 a month in the US | More messages with tools, uploads, images and memory; may show ads |
| Plus | $20 a month | Reasoning models, deep research, projects, scheduled tasks, custom GPTs, Codex on the Sol and Astra models |
| Pro | From $100 a month, three usage tiers | Pro reasoning, longer Codex sessions, no five-hour limit in Codex for now |
From chatgpt.com/plans/plus, chatgpt.com/pricing, OpenAI’s Help Center, its Codex pricing page and its Go launch post, read on 22 November 2026. Prices before tax.
Two details most price guides skip. Plus does not include the API: OpenAI bills API keys separately. And the meter itself is about to change.
From December, ChatGPT Plus runs on one token meter
OpenAI’s help page on ChatGPT and Codex usage limits, updated on 22 November 2026, says that from December Plus, Pro and Business share one allowance across ChatGPT and Codex, counted in tokens. Bigger models and harder tasks use more of it. Instant chat will not draw from it, and the remaining share and reset times move to Settings → Usage.
So “is Plus worth it” becomes “how much of that meter will my week use”. Codex already works this way today: a five-hour window and a weekly limit, each shown as a percentage. That is the meter we measured.
What a heavy week uses on ChatGPT Plus
Our Plus seat does agent work: coding jobs through Codex on GPT-6 Astra and GPT-6.1 Sol, often several at once. Codex writes the tokens of every turn into its log next to the meter’s reading, so you can see exactly how much work moves the meter one point.

- A five-hour window is about 15% of the week. We ran five windows to their limit, and each one took between 14% and 16% of the week. About seven full windows empty it.
- On the top model, a window lasts minutes. Agent jobs on GPT-6 Astra filled the window in 22 to 37 minutes. With mostly GPT-6.1 Sol, the slowest window lasted 93 minutes.
- One day can take half the week. On 29 October three windows ran dry back to back, and the week’s meter went from 1% to 54% in less than a day.
Now put a price on it. OpenAI lists credit rates for Codex per million input, cached and output tokens, and its dollar prices on the same page work out to four cents a credit. At those rates, one point of the week’s meter held between $0.58 and $0.93 of work, $0.72 on average. A full week is about $72, and a month of weeks about $310. That is the ceiling, though: you only get it if you fill every week.
What Free would have cut
For chat, Free cuts less than you’d think: it includes unlimited text chats with GPT-6. What it cuts is the heavy work. On Free, Codex runs only GPT-6 Luna, in the desktop app. Every one of the 904 model calls in our Codex logs ran on Sol or Astra, so none of that week could have run on Free.
Free also limits uploads, image creation, voice and deep research. OpenAI doesn’t publish those caps and we didn’t measure them. If what you need is more chat with files and pictures, and not the reasoning models, Go at $8 is the cheaper step up.
When ChatGPT Plus stops being enough
Over four months our Plus week ran out three times, the earliest 2.4 days before it reset. When that happens you can wait, buy credits, or move to Pro. A turn already running gets to finish.
Before paying five times more, change the model. OpenAI’s own Codex table gives Plus 17 to 180 local messages per five hours on GPT-6.2 Sol and 350 to 3,000 on GPT-6 Luna, about 17 to 20 times more. Send the easy work to the small model and keep Sol or Astra for the hard part, and the same week lasts much longer. Pro starts making sense when even that runs dry, or when you need long sessions with no five-hour stop.
Is ChatGPT Plus worth it for you? Test it on your own week
You don’t need our numbers for this, only your own meter. In Codex, /status shows it; on the web, the Codex usage dashboard; from December, Settings → Usage in ChatGPT.
Read the week’s percentage the day after it resets, read it again three or four days later, and divide the difference by the days. That gives your daily pace, and two lines tell you what it means:
- Under about 1% a day, Plus costs more than your week is worth. $20 a month is $4.62 a week, and at our $0.72 a point that buys about 6.4 points. If you also never pick a reasoning model, try a week on Free or Go.
- Above about 14% a day, you will run out before the reset (100% over seven days). Move routine work to a smaller model first; then consider credits or Pro.
Anything between the two lines is the zone where $20 buys more work than it costs. Your model mix will move our $0.72 a point up or down, so treat both lines as a starting point.
FAQ
How much is ChatGPT Plus per month?
$20 a month in the US, billed monthly, before tax. Go costs $8 a month in the US and Pro starts at $100.
Is there a yearly ChatGPT Plus plan?
No. OpenAI’s Help Center says Plus has no annual billing and can’t be paid several months in advance.
Does ChatGPT Plus have usage limits?
Yes. Codex and agent work have a five-hour window and a weekly limit. From December, ChatGPT and Codex share one token allowance on Plus, and Instant chat does not count against it. On our seat, one full five-hour window was about 15% of the week.
What happens when you hit the ChatGPT Plus limit?
You wait for the reset, buy credits, or upgrade. If you reach the limit in the middle of a task, that turn is allowed to finish.
Is ChatGPT Plus worth it compared with the free plan?
For chat alone, often not: Free includes unlimited text chats with GPT-6. Plus pays off once you use reasoning models, deep research or Codex. By our numbers it beats paying per token from about one point of the weekly meter a day.
The meter on screen, in TinkerClaw
We could only measure this because our agent runs on TinkerClaw, which keeps every plan’s meter in view while it works. It reads the ChatGPT five-hour and weekly meters next to Claude’s, so a plan about to run dry is visible before it stops a job. See under the hood walks through that panel, and its Auto setting sends easy work to small models, the same trick that stretches a Plus week; the post on Auto explains how it picks. You can click through all of it in the live demo.
The same tools come as skills on ClawHub: the Token Panel for every provider’s spend and the Model Router, which sends work to flat-rate plans first. For the Claude side of the same question, Claude Code pricing measures how much work Pro, Max and Ultra hold, and OpenClaw cost per month puts the subscriptions into a full agent bill. If you are weighing a subscription against your own hardware, AI costs in 2026 runs that comparison.
For builders
- Budget Prompting (PDF): why every turn re-sends its context and how to cut what that costs; the paper behind What a heavy week uses on ChatGPT Plus.
- There are others, by topic, in the brain map and the site map.

Leave a Reply