Why one AI subscription beats ChatGPT alone

ChatGPT Plus is a good subscription. That is a more honest start than most pieces on this topic make, because for €21.02 per month you get a lot: GPT-5.6 Sol as the top model, 160 messages per three hours, Sora 2 for video, GPT Image 2 for images, Codex, Agent Mode and Deep Research. If you use one AI tool, this is one of the better choices you can make.

The problem isn't ChatGPT. It is the second subscription. Because almost nobody sticks with one model: you notice that Claude writes nicer Dutch, that Gemini handles thick PDFs better, that Perplexity shows its sources. So you add one. And with that your monthly cost goes from €21.02 to €42.04, and with three subscriptions to €64.03.

This piece is about that math, and about when bundling does and doesn't make sense. Including the cases where you should stay with one provider.

What you get from ChatGPT for €21.02

Let's first be precise about what is in the subscription, because that often gets glossed over in comparisons like this.

ChatGPT Plus gives access to GPT-5.6 Sol, the flagship from 9 July 2026, with a context window of 1.05 million tokens. The 6 August 2026 update brought better factual accuracy and an effort slider that lets you decide how long the model thinks. You get 160 messages per three hours on the top model. On top of that, Sora 2 and GPT Image 2 are included for video and images, plus Codex, Agent Mode and Deep Research.

That package is broader than what the competition delivers for the same money. Claude Pro has no video, Google AI Pro has no Sora equivalent in the same price range and Perplexity is an answer engine, not a creative package. For anyone who makes images and video, ChatGPT Plus is not expensive but actually cheap.

And the free version has become serious. Since 6 August 2026, free users can text chat without limits on GPT-5.6 Luna with a Think button; the limits still apply to uploads, images and tools. For some of the people paying for Plus now, that is enough.

The problem starts with the second model

The models are good at different things, and that is not a marketing story but measurable. Claude scores a 5 on Dutch in our language test against a 4 for ChatGPT. Gemini 3.1 Pro, with NotebookLM and 1 million tokens of context, is the strongest for document work. Perplexity is the only one that puts sources with every answer by default. Grok 4.5 is strong at code and has live data from X.

Anyone who works seriously with AI discovers this within a few months and starts switching. That is sensible behavior. The market just punishes it, because every model sits behind its own paywall.

With three subscriptions you are at €768.36 per year. That is more than many households pay for phone, streaming and internet combined, for tools you probably don't fully use: you don't use Sora every day, you don't ask for sources every day, and Claude Code sits idle if you don't code.

The detailed math, including what you actually use per subscription, is in why multiple AI subscriptions are a waste of money.

Why stacking happens so easily

Nobody decides one day to spend €64 a month on AI. It happens in steps of €21, and each step is reasonably justifiable on its own.

You start with ChatGPT Plus. Two months later you have a big text that needs good Dutch and you add Claude Pro, "just for this project". Then a client comes along with a two-hundred-page report and Google AI Pro turns out to handle it better. Each subscription solves a real problem, and none of the three gets cancelled, because the monthly amount is just low enough not to stand out.

The point where it hurts is usually the annual statement or the moment an accountant brings it up. Then it turns out one of the three hasn't been used in months.

What a bundle solves, and what it doesn't

There are two bundle routes, and it is only fair to name both with their downsides.

Poe is the American route. Quora bundles dozens of models behind one subscription and bills in compute points. Premium costs $19.99 (€21.01) for 660,000 points per month, Starter $4.99 (€5.25) for 10,000 points per day. Technically that works well, but it has two practical downsides. The points system is unpredictable: a question to Claude Fable 5 costs dozens of times more points than the same question to a cheap model, points don't roll over and buying extra costs about $30 per million points. On top of that, Poe cut back in 2026: the free tier went from about 3,000 to about 300 points per day and Premium from 1,000,000 to 660,000 points per month. And everything around it is in English, including invoices and customer service; in our language rating the app therefore scores a 3, while the models in it speak fine Dutch.

We are the Dutch route. The same idea, but with fixed, understandable limits instead of points, a Dutch interface and invoices in euros including VAT. Plus costs €12.95 per month or €129 per year, Pro €29.95 per month for heavy daily use and agent tasks, and there is a free tier with a daily message limit on the fast models, without a credit card.

What a bundle doesn't solve, and we have to say this just as clearly: the provider-specific extras stay with the provider. Sora 2 belongs to OpenAI. Claude Code and Claude Cowork belong to Claude Pro. NotebookLM and the 2 TB of storage belong to Google AI Pro. A bundle gives you the models, not each vendor's surrounding programs. If you need exactly one of those extras every day, just keep that subscription.

When you should stay with one provider

This is the part a sales piece would leave out. In four cases a single provider is the better choice.

You code every day. At €21.02, Claude Pro is the only subscription in this class with a full coding environment included. If you want to stay in your editor, you end up at GitHub Copilot Pro for $10 (€10.51) or Cursor for $20 (€21.02). See coding with AI.

You make video or images. Sora 2 and GPT Image 2 are in ChatGPT Plus and in no other subscription in this price range.

You spend all day in Office. Microsoft 365 Premium ($19.99, about €21) gives you Copilot in Word and Excel plus the Office apps themselves. As a bundle that is unbeatable, because the Office license is included.

You are price-sensitive and really use one model. Then Google AI Plus at €7.99 per month is the cheapest serious choice, and any bundle costs more than necessary.

A direct comparison of the three plans around €21 is in ChatGPT Plus vs Claude Pro vs Google AI Pro, and which model fits which task in GPT-5.6 vs Claude 5 vs Gemini 3.5.

Where the market is heading

Two developments make stacking subscriptions harder and harder to defend.

The free tiers have become good. Free ChatGPT gives unlimited text chat on GPT-5.6 Luna, and free Claude runs on Sonnet 5, a top model. For light users the question is no longer which subscription, but whether a subscription is needed at all.

And the mid-range has collapsed. GPT-5.6 Terra is roughly as strong as last year's flagship, but in July 2026 OpenAI cut the price of Terra by 20% and of Luna by 80%. Models you paid €21 a month for last year are now almost free. The cost is only in the newest generation, and you don't need that for most tasks.

What that means in the longer term is discussed in the future of AI subscriptions. If you want to see the full price comparison first, go to our comparison page or straight to our pricing.

Conclusion