ChatGPT Plus vs paying per token: what a month of AI really costs
· 4 min read
ChatGPT Plus costs $20 a month. In Europe VAT comes on top, so expect roughly €22 to €24 depending on your country. You pay that in a month where you ask it two questions and in a month where you live in it. The API that powers these models works the other way round: you pay for the text you send and get back, and nothing else.
So which one is cheaper? It depends almost entirely on how much you use it and which model you pick. Below are the numbers for three kinds of users, with prices taken from the OpenAI and Anthropic pricing pages on 19 October 2026.
How per-token pricing works
Models read and write in tokens. A token is a chunk of a word; 1,000 tokens is roughly 750 English words. Providers charge per million tokens, with a separate price for input (what you send) and output (what the model writes).
The part most people miss: in a chat, every new message sends the whole conversation so far back to the model. A 20-message thread isn't 20 small requests. The last message carries everything above it. Long chats are where per-token costs grow, so starting a fresh chat for a new topic is the single best habit if you pay this way.
What the models cost right now
| Tier | Examples | Input, per 1M tokens | Output, per 1M tokens |
|---|---|---|---|
| Small | GPT-6 Luna, Claude Haiku 5.5 | $0.10 | $0.50 |
| Mid | GPT-6.1 Sol, Claude Sonnet 5.5 | $2 | $10 |
| Top | GPT-6 Astra, Claude Fable 5.1 | $10 | $50 |
The mid tier is what most people actually need for writing, summarising and everyday questions. The small tier is fine for quick lookups, rewording and translation. The top tier is for hard reasoning and long, careful work, and it costs five times as much as the mid tier.
Three users, three bills
These are estimates, not promises. We assumed conversations get longer as they go, so the average message carries a good amount of history.
| User | Usage per month | Small | Mid | Top |
|---|---|---|---|---|
| Light | About 5 chats a week, 5 messages each (100 messages, 200k tokens in, 50k out) | $0.05 | $0.90 | $4.50 |
| Regular | 3 chats a day, 8 messages each (720 messages, 2.9M tokens in, 0.5M out) | $0.54 | $10.80 | $54 |
| Heavy | 10 long work chats every working day, pasting documents (3,300 messages, 33M tokens in, 3.3M out) | $4.95 | $99 | $495 |
A few things stand out.
- If you're a light user, Plus is mostly money for nothing. Even on the most expensive models you'd spend less than a quarter of the subscription.
- A regular user on a mid-tier model pays about half of what Plus costs. Use the top models for everything and the subscription wins.
- Heavy users should stay on a subscription, or keep the expensive models for the jobs that need them. The flat price exists for exactly this kind of use.
Prompt caching changes the picture in your favour. Providers charge much less for input they've seen recently: OpenAI lists cached input for GPT-6.1 Sol at $0.10 per million tokens instead of $2. Not every chat app takes advantage of it, so it's worth checking before you buy one.
What you give up
Plus isn't only a model. It bundles image generation, voice mode, deep research, projects and other features into the same $20. With the API you get the model and build the rest yourself, or pay for an app that does. If you use those extras every day, the subscription is better value than the token maths suggests.
How to switch to paying per token
- Create an API account and set a spending limit. Both OpenAI and Anthropic let you prepay or cap your monthly spend, so a forgotten script can't run up a bill. Start low; you can always raise it.
-
Pick an app to chat in. The provider consoles work, but they're built for developers. A desktop app
that takes your API key is nicer to live with, and some of them are a one-time purchase:
- TypingMind sells lifetime licences for individuals, starting at $39. You add your own keys and pay the providers directly.
- BoltAI is a Mac and Windows app with perpetual licences from $79, including a year of updates. You keep the last version if you don't renew, but cloud sync and mobile need a renewal.
- Or use one account for every model. OpenRouter gives you a single balance for models from many providers at the providers' own token prices. It takes a 5.5% fee when you buy credits by card.
- Check your usage after the first month. The dashboard shows exactly what you spent. If it's well under $20, you have your answer. If it's creeping above, look at which model and which chats are doing it.
The short answer
If you use ChatGPT a few times a day for ordinary work, paying per token on a mid-tier model will usually cost you about half of Plus, often less. If AI is open on your screen all day and you want the best model every time, keep the subscription. Most people are closer to the first group than they think, and the first month of API usage will tell you which one you are.
More ways to pay for AI by use are in our AI & Chat category. For other kinds of software, see software without a subscription.