AI / GenAI·6 min·15 September 2026·Which AI assistant? — Part 1 of 6

Which AI assistant? Where free ends and paid begins

The question I get most often isn’t about prompts. It’s about money. “I use ChatGPT for free, should I pay? Is Claude better? And what even is Gemini?” Usually from someone who types something in now and then and wants to know whether they’re missing out. This is the first of six parts answering that question for five assistants: ChatGPT, Claude, Gemini, Grok and Perplexity. Each part takes one question and puts the five side by side. This one covers the basics: where does the free version end and the paid one begin?

First, what you should know about me, because it colours the piece. I pay for Claude, on the heaviest plan an individual can get. I pay for Gemini, on the lightest. ChatGPT and Perplexity I use for free. Grok I don’t use as a chatbot at all, only as a news robot on X. So on the paid versions of ChatGPT, Perplexity and Grok, all I can tell you is what their makers write about them.

What free actually means

None of the five treats free as a trial. You can keep using it indefinitely. The difference sits in three things you only notice once you hit them.

The first is the model. Every maker has a fast, small model and a heavy one that reasons better. Free gets you mostly the small one, and the heavy one in small doses. ChatGPT says it plainest: unlimited chat with the light model, and “limited access” to the rest.

The second is how much it can hold in one conversation. That’s called the context window, and it’s why a free assistant loses the thread halfway through a long chat. ChatGPT publishes the numbers: about twelve pages of text as input on free, forty on paid, two hundred and fifty on the most expensive tier. The others don’t give a figure, but the pattern is the same.

The third is what comes around it: uploading files, a project with standing instructions, a memory that carries across chats, a connection to your mail. Part 3 is about that. For now: Claude puts most of it in the free tier, ChatGPT the least.

The table

Checked on 6 September 2026 from the Netherlands, monthly prices, excluding VAT where the page said so. Prices in this market don’t survive a quarter; when something changes I’ll update the table, and you’ll see it in the date above the piece.

What free gets youFirst paid stepHeaviest individual tier
ChatGPTUnlimited chat with the light model, few uploads, little memory, no projectsGo, 8 euros: more uploads, longer memory, possibly ads. Plus, 23 euros: the heavy model, projects, your own variants of the assistantPro, from 103 euros
ClaudeWeb search, memory, projects, reusable working instructions and one custom connection; but a tight usage limitPro, 18 euros (15 billed annually): more usage, more models, Claude CodeMax, from 90 euros: five to twenty times the usage, earlier access to new features
GeminiThe app with the fast model, the heavy one in limited dosesPlus, 4.99 euros: double the limit, 100 GB of storage, sharing with your household. Pro, 21.99 euros: four times the limit, more of the heavy modelUltra, 99.99 or 219.99 euros
GrokChat on X and grok.com, search across the web and X, memoryX Premium, 8 dollars: higher limit. Premium+, 40 dollarsSuperGrok; the price isn’t on the site, so I’m not quoting one
PerplexityUnlimited basic search, three uses a day of the heavy search, one deep research a month, no model choicePro, 17 dollars billed annually: model choice, research, uploadsMax, 167 dollars billed annually

Two things stand out. The tier where you actually get the heavy model sits around twenty euros at ChatGPT, Claude, Gemini and Perplexity, and below it ChatGPT and Gemini each have a cheap tier that mostly sells storage and looser limits. And the top end is a different product at all five: a hundred euros a month or more is for people who do their job in the tool, not for the question this piece opened with.

What I see myself

My experience with free ChatGPT is short: you can’t do much with it beyond looking things up and small bits of text. That isn’t a verdict on the model, it’s what the free tier allows.

I pay the most for Claude, and that didn’t start with a limit. I was using it for free and that felt odd, because you could see how much work had gone into it. The subscription was support first and access second. The limits came later: I ran into them on the middle tier, and I still do now and then on the heaviest one. What you buy on top is speed, more usage and earlier access to new features.

I have Gemini on the lightest paid tier, honestly just to try it. Storage was the argument, along with sharing it with someone at home and the fact that it sits inside every Google app, on the iPhone too. Results so far are good, especially for looking things up and for images and video.

I use Perplexity for free, and I hit exactly the three heavy searches a day the table mentions. The reason it hasn’t turned into a subscription won’t apply to most readers: I have separate credit for the developer interface, and I use that from Claude Code.

I only know Grok from three automations on X that send me wine and drinks news every day, two international and one Dutch. Part 2 comes back to those.

Caveat: the price isn’t the amount

What you pay is in the table. What you hand over isn’t. ChatGPT states on its own pricing page that your conversations are used to train the model, with a switch to turn that off, on every tier including the paid ones. Claude states on its pricing page that this doesn’t happen by default. At Gemini and Perplexity it sits behind a setting you have to find yourself, and what the default is there I’ll work out for part 4. I didn’t know that switch existed until I prepared this piece, and I work with these tools every day. For now the caveat under every price in the table is this: the amount is the part of the bill you can see.

The second caveat is dependence. In this piece I can’t say anything from experience about the paid side of three tools. That’s no accident. Once you pay and your projects, memory and connections all live in one assistant, you don’t just switch. The table compares prices; what you can still compare after that gets smaller every year.

What to do tomorrow

Look in the assistant you already use and check which model it turns on for you. In ChatGPT it’s at the top of the chat, in Claude and Gemini it’s in the menu next to the input field. If the light model is sitting there and you regularly ask it things it needs to think about, you now know where the difference you’ve been feeling comes from.

Frequently asked questions

Is the free version good enough to start with?

Yes, at all five. Start free and watch for the moment you want something it won’t do: a long file, a conversation that loses the thread, a limit you hit halfway through the day. That moment tells you which paid tier you need, and whether you need one at all.

Can I use two free versions side by side?

Yes, and for someone who types something in now and then, that often beats one subscription. Part 6 of this series is about that.


Sources: Claude pricing, ChatGPT pricing, Gemini subscriptions, Perplexity Pro, Perplexity plans, Grok, X Premium, model training at Claude, ChatGPT data controls, Gemini apps privacy. Checked 6 September 2026.