ChatGPT Plus vs Pro: is Pro worth it? We ran 54 coding jobs to find out

ChatGPT Plus costs $20 a month. ChatGPT Pro now comes in three sizes: $100, $200 and $500. OpenAI does not say how much more you can do with Pro. So we measured it on the site owner's own Pro 200 plan: we let Codex, OpenAI's coding tool, do 54 small coding jobs. All 54 jobs passed every hidden check, and the weekly counter only moved from 7% to 8%. By our count, more than 4,800 jobs like ours fit in one week on Pro 200.

What do Plus and Pro cost now?

On OpenAI's help and pricing pages, in US dollars:

• Free: $0. For "quick coding tasks" in Codex, OpenAI's Codex pricing page says.
• Go: $8 a month. More messages, uploads and images than Free, on the pricing page. OpenAI says it may show ads.
• Plus: $20 a month, billed monthly.
• Pro: three plans: Pro 100 for $100, Pro 200 for $200 and Pro 500 for $500 a month. Only Pro 500 has Ultrafast, the fastest mode. Pro is billed monthly only.

Prices outside the US can differ. On the Dutch version of the pricing page, Plus was €23 and Pro started at €103 a month.

What do you get with Plus and Pro?

Plus gives you OpenAI's reasoning models with GPT-6, more messages and uploads, deeper research, projects, scheduled tasks and custom GPTs. It also gives you more use of Codex, OpenAI's coding agent.

Pro has everything in Plus. It adds Pro reasoning with GPT-6 Astra and longer sessions in Codex. It also adds Dot, an agent that is always on, unlimited image creation and the most memory. The biggest difference for coders is the limit. On the Codex pricing page, Plus has a 5-hour limit. OpenAI estimates, for example, 15 to 160 messages per 5 hours with GPT-6.1 Sol, depending on how big the tasks are. A weekly limit may also apply. Pro "currently" has no 5-hour limit.

But OpenAI gives no numbers for Pro's weekly limit, and no number for how much more Pro gives than Plus. That is why we measured it.

Our test

We used the same three small coding jobs as in our Claude Pro vs Max test. They are: fix the money maths in a small program, write a bill splitter, and write a CSV reader. Each job ran in its own folder, and hidden tests checked the code afterwards. More about this in how we test AI models.

We ran each job 18 times, 54 jobs in total, with Codex and the GPT-6.1 Sol model at medium effort. Codex is OpenAI's coding tool for the command line, also called the CLI. It ran in a clean setup with only the owner's ChatGPT Pro 200 login: no personal settings or add-ons. Nobody else used Codex on that account during the test. A typical job took about 110 seconds.

The result: all 54 jobs passed every hidden check, 1,116 checks in total.

What Codex reports

While it works, Codex writes how much of the plan's limit is used into its own log. On our Pro 200 plan, it showed one window of 10,080 minutes, which is one week. It showed no 5-hour window, which matches OpenAI's page.

The number is a whole percent, like 7% or 8%. So one small job does not show up on its own. That is why we ran many jobs and watched when the number went up.

How much one coding job uses on Pro 200

At the start, the weekly limit was 7% used. During job 6, it went to 8%. Then it stayed at 8% for all 48 jobs after that.

So those 48 jobs together used less than one point of the counter. That means one job used on average less than 0.021% of the week, and more than 4,800 jobs like ours fit in one week. That is a lower limit: the real number can be higher, because the next step up did not come.

Our jobs were small, about two minutes each. A big project with a long chat uses much more per step. OpenAI's help page also says Fast mode uses the limit 2.5 times as fast.

What fits in a week, and on Plus

Here is how that compares, for small jobs like ours:

• ChatGPT Pro 200, our test: more than 4,800 jobs a week.
• ChatGPT Plus, OpenAI's estimate: 15 to 160 GPT-6.1 Sol messages per 5 hours, plus a weekly limit without numbers. OpenAI says bigger tasks use more.
• Claude Max 5x ($100), our earlier test: roughly 1,400 small jobs a week, somewhere between 900 and 2,700, measured over three Claude models.
• Claude Pro ($20), our estimate from that test: about 90 Sonnet jobs per 5-hour limit, somewhere between 60 and 180.

These are different tools, models and plan prices, and the Claude jobs were shorter. Our ChatGPT number is a lower limit; the Claude numbers are middle estimates. So this is only a rough guide, not a fair race. It does show that Pro 200 gave us a lot of room for small coding jobs.

Is Pro 200 good value?

Each job read about 94,356 tokens and wrote about 2,898. 89% of what it read came from the cache: text the model had already seen, which is much cheaper. A token is a short piece of a word. The API is the way developers pay per token instead of per month. At OpenAI's API prices for GPT-6.1 Sol, that is about 5.8 cents per job. Our 54 jobs would have cost about $3.14.

If you ran every job our test says fits, 4,800 a week, the API would cost at least about $1,112 for four weeks. Pro 200 costs $200. So for heavy daily coding, Pro 200 can be much cheaper than the API.

But only if you really use it. Most people will not run thousands of coding jobs a week. If you code now and then, Plus or the API may be cheaper. See our AI API price comparison.

The Pro 200 change on 29 October

OpenAI has changed Pro 200. Its help page says new Pro 200 plans include "a lower usage allowance than previously offered", for the same $200.

People who had Pro 200 at some point between 22 and 29 September keep the old, higher allowance through 29 October. After that, they move to the lower one too. OpenAI says only those people get an email.

The site owner got no such email, so our test plan most likely already had the new, lower allowance. Our weekly limit also resets on 29 October. We may test again after that and add a dated update here.

Which plan should you pick?

• You chat, write and code now and then? Plus is enough. It has the reasoning models and some Codex.
• You code with Codex every day and hit Plus's limits? Pro 100 or Pro 200. Neither currently has a 5-hour limit. We only measured Pro 200.
• You need the fastest answers? Pro 500 is the only plan with Ultrafast.
• You just want to try it out? Start with Free or Go, and upgrade when you hit a limit.

For a comparison of the tools themselves, see Codex vs Claude Code.

How we checked

From 21 to 23 October 2026 we read OpenAI's help pages on Plus and on the Pro plans. We also read the Codex pricing page and the ChatGPT pricing page. The ChatGPT pricing page showed us its Dutch version, with prices in euros.

On 23 October, between 13:18 and 14:58 UTC, we ran 54 small coding jobs on the site owner's ChatGPT Pro 200 plan. We used Codex CLI 0.163 with GPT-6.1 Sol at medium effort, in a clean setup with only the ChatGPT login. After every job we read the rate-limit lines from Codex's own session log, and hidden tests checked the code.

We did not test Plus, Pro 100 or Pro 500, because we do not buy or change plans for a test. Numbers for those come from OpenAI's pages. The API cost uses OpenAI's list price for GPT-6.1 Sol. The Claude numbers come from our test of 12 October. The charts are our own, and none of the images was made by AI. One plan, one model and small jobs: your use may differ.