Which AI to pay for: the subscription that pays off in 2026

After «which AI should I use», the question I get most is which AI to pay for. And here almost everything depends on how much you really use it.

ChatGPT, Claude and Gemini prices in Spain (September 2026)

Prices checked on 24 September 2026 on each company’s website, visiting from Spain. OpenAI, Google and Microsoft already show them with VAT. Anthropic shows them without VAT, which is why I give you both figures, and Perplexity charges in dollars and adds the tax at checkout. They change often, so check before you pay.

Is it worth paying? Free, €20 or €100 and up

• Free, if you use it now and then: a question, an email, a summary. The free versions of ChatGPT, Claude and Gemini are already very good.
• About €20 a month, as a next step if the free plan falls short on a useful task. Claude Pro includes Claude Code and ChatGPT Plus expands Codex and ChatGPT Work, but available models and limits depend on the plan. Paying does not guarantee the best result or uninterrupted use.
• €100 a month or more, when you have verified a need for more capacity or a specific feature and the cost makes sense. Check limits and possible extra charges before upgrading.

On that top step are Claude Max (5 or 20 times the usage of Pro), Google AI Ultra and ChatGPT Pro. Update on 30 September 2026: OpenAI lists three tiers, Pro 100, 200 and 500, at published prices of $100, $200 and $500 a month. Pro 200 is open to new subscriptions again; new subscriptions include less usage than the previous version. Eligible existing subscribers keep the previous allowance temporarily until 29 October. Ultrafast belongs to Pro 500. These are dollar references, not euro conversions: confirm the total for your account before changing plans.

My advice: pay for one. If you are torn between two, keep the one that fits the tools you already use.

ChatGPT Go or Plus: which one suits you

Go costs €8 and Plus €23, both with VAT. You notice that €15 gap in three places:

• The models. Plus broadens reasoning access compared with Go. OpenAI says availability changes during rollouts: check your account’s picker and access in Chat, Work and Codex separately. Do not assume they all offer the same model or allowance.
• Agents. With Go, as with the free plan, you only get a limited taste of Codex and ChatGPT Work, and Work only in the desktop app. With Plus you get much more use of both, and Work on the web and your phone as well.
• Ads. OpenAI warns that Go may show ads. Plus has none.

Go does give you more room than the free plan for images, files and memory. For writing emails and asking questions, you can start free within your account limits. Consider Go if you run short uploading files or making images. If a useful task needs more capacity or a specific model, check that the plan includes it before paying. A contract still needs your review and, where appropriate, a professional’s.

ChatGPT Plus or Claude Pro

Which one is better depends on the task, and I go through that properly in which AI to use for what. Here is what you get for your money, because they cost almost the same:

• ChatGPT Plus, €23. The most complete app, with images and, for my taste, the best voice mode. It comes with Codex, the agent that works on your computer, and ChatGPT Work for long office jobs. Check the models and limits available in each.
• Claude Pro, €21.78. Opus 5.5, which as of 24 September 2026 is top of the independent Artificial Analysis ranking. It comes with Claude Code to work on your computer and Cowork, its office agent, which Anthropic is folding into the normal chat.

If what you are after is an agent for office work, I compare the two in Claude Cowork vs ChatGPT Work.

And if you work in Gmail and Drive, there is a third way: Google AI Pro (in Spanish), at €21.99, with Gemini inside Gmail and Docs and 5 TB of storage.

Monthly or yearly

Paying for the year upfront is cheaper, but not all of them let you. This is how it stands on 24 September 2026:

• ChatGPT. Go, Plus and Pro are monthly only. Yearly billing only exists on the business plans.
• Claude. Pro costs €180 + VAT a year (€217.80), about €44 less than paying monthly. Max is monthly only.
• Google AI Pro. €219.99 a year through Google One, against €263.88 paying monthly.
• Microsoft 365 Personal, the one with Copilot: €99 a year or €10 a month.
• Perplexity Pro. $17 a month if you pay for the year, plus VAT.

I would start monthly. In AI the leader changes every few weeks, and this way you can move without losing money. Go yearly only when you have used it every day for a few months and you know you are staying.

Usage limits, explained

The fee includes limited usage; Claude explains that the model, files, conversation and tools affect it. Two tasks can consume very different amounts. Reset windows and restrictions depend on the product and account: check Usage. Plus does not include API usage either. Credits or extra usage can cost more; check spending limits before enabling them.

Now and then the companies give away a reset, a button that refills your limit in one go. When Opus 5.5 came out on 22 September 2026, for example, Anthropic gave one to Pro, Max and Team plans that expires on 22 October 2026. The same day, with GPT-6 Sol and Luna, OpenAI left another one banked in Codex for Plus, Pro and Business. Claude’s is under Settings, Usage, with its expiry date. It can only be pressed from the web or the desktop app, but it also works for Claude Code, because the limit is shared. If you have them, save them for the day you run out halfway through something important.

How to work out which one pays off

For a week, write down the tasks you do and how long each takes. Then ask the AI itself:

These are all my tasks and measured times for a week: [without AI in comparable cases, preparation and AI use, review and corrections, acceptable or pending result]. Include failed attempts and work redone manually, without counting the same time twice. If tasks remain unfinished or data is incomplete, do not present the subtotal as overall savings. Propose a small trial for anything missing. Separate observations from estimates; do not invent savings or quality. Compare the free plan with a paid plan using current prices and limits, including tax and extras. If data is missing, say what to measure before recommending payment.

Saving two hours does not guarantee a return. Count preparation, review, failed attempts and work redone manually. One task saving ten minutes and another taking fifteen extra minutes leave five minutes lost overall. Compare the value of your time with the subscription, unrecoverable taxes and extras. You can save time without earning more. If data is missing, the sample is small or quality changes, keep measuring before treating the saving as established.

Before you put work data in

On personal ChatGPT, Claude and Gemini accounts, check whether you have allowed your conversations to be used for training. The setting is not the same everywhere: Claude makes it depend on your consent. Turn it off if enabled; doing so does not by itself authorize the use of client data. For serious work with other people’s personal data, use business accounts such as ChatGPT Business or Claude Team, which do not train on your data by default, and check with whoever handles data protection.

Sources: savings measurement reviewed on 2 October 2026; Pro tiers and limits on 30 September. Local prices in the table retain the 24 September check; they are not a guaranteed current offer. Other references checked on 26 September: ChatGPT prices, Claude prices (in Spanish), Google One plans and Google AI plans (in Spanish), Microsoft 365 plans (in Spanish), ChatGPT models by plan, ChatGPT Work and Codex by plan, what ChatGPT Plus includes, what ChatGPT Go includes, ChatGPT Pro tiers, Perplexity Pro, how Claude resets work and ChatGPT Business data.