Which AI Assistant Is Worth Paying For?

Three companies arrived at the same price. ChatGPT Plus is $20 a month. Claude Pro is $20 a month, or $17 if you pay for the year up front. Google AI Pro is $19.99. That is not a coincidence and it is not a coordination scandal either — it is what happens when everyone works out what a consumer will pay for one subscription and stops there.

Which is convenient for them and awkward for you, because it removes the one comparison people are good at making. You cannot pick the cheap one. There isn't a cheap one.

What the money actually buys at each tier

Published rates at the time of writing, in the currency each vendor quotes.

Plan	Monthly	Annual equivalent	What it adds
ChatGPT Plus	$20	—	Higher limits than free, the full feature set
ChatGPT Pro	$100 / $200	—	5x and 20x the usage of Plus respectively
Claude Pro	$20	$17/mo, $200 up front	Claude Code, unlimited projects, more usage
Claude Max	From $100	—	5x or 20x Pro usage, priority at busy times
Google AI Plus	$4.99	—	400GB storage, modest usage lift
Google AI Pro	$19.99	—	5TB storage, higher limits, Deep Research
Google AI Ultra	$99.99 / $199.99	—	20TB storage, 5x and 20x limits

Two things stand out. Google's $4.99 tier has no equivalent anywhere else, and it is the only genuinely cheap way into any of this. And every step above $20 is sold as a multiplier rather than a number, which tells you the vendors themselves cannot promise a fixed allowance.

Pick on the work, not the model

Model rankings turn over every few months. What you do all day does not. So sort yourself into one of these before you look at a benchmark.

• You work in long documents. Contracts, reports, research papers, a codebase. Claude has been the steadier choice here for a while, and Pro includes Claude Code, which means it can work on files rather than talk about them.
• You want one bill for everything. Text, images, voice, browsing, agents. ChatGPT has the broadest surface area, and it is the one every tutorial and template assumes.
• You live in Google's apps. If your day is Gmail, Docs and Drive, the assistant that is already inside them wins on friction alone, and the 5TB removes a separate line from your bill.

Notice that none of those three sentences mentions which model scores highest on anything. That is deliberate. At this price point the differences that survive a month of real use are about where the tool sits in your day, not about a leaderboard.

The tiers above $20 are a different product

It is easy to read $100 as "the good one" and $20 as "the trial". That is not what is happening. The $100 and $200 tiers exist for people who hit limits on the $20 plan repeatedly, and the honest test is whether you have.

Both OpenAI and Anthropic describe their upper tiers in multiples: 5x and 20x. Google does the same with Ultra. Nobody publishes the base number those multiply, so you cannot work out in advance whether 5x is enough — you can only find out by running out at 1x first. Which is the correct order to do it in anyway.

If you have never seen a usage limit message, you are on the right plan. If you see one most days, the upgrade will pay for itself in the time you currently spend waiting for a reset.

What the free tiers are actually for

All three publish a free plan, and they are not demos. For someone who opens an assistant a few times a week to draft an email or check a fact, the free plan is genuinely the right answer, and a subscription would be money spent on capability you never reach.

The useful thing about the free tiers is that they let you run the comparison properly. Take one real task from your week — not a puzzle, not a benchmark question, the actual thing — and give it to all three. The differences show up immediately on real work and almost never on test prompts. Where the answers matter enough to check, the same habits apply as with any tool that produces confident text: our piece on AI research tools and your sources covers what to verify and how.

The cost nobody puts in the comparison

The subscription is $20. The real cost is that you now have a habit, and habits attract other subscriptions. We have watched teams end up paying for two assistants, a writing tool built on one of them, and a meeting notetaker built on another — four bills, three of which are wrappers around the model in the first bill.

Before you add the second AI subscription, check whether the one you have already does the job. It usually does, slightly worse, for free.

What we would do

Start free for two weeks with a real task, not a trial task. Then pay for one, monthly rather than annually, so that a change of mind costs you one month rather than a year. Put a reminder in the calendar for ninety days out asking one question: what did this replace? If the answer is nothing, cancel it.

And if you already pay Google for storage, run that sum first. It is the only part of this comparison where the answer is arithmetic rather than judgement, and arithmetic is easier to be right about.