AI tokens, explained: why the same text can cost more on a new model
AI companies charge by the token, not by the word. Here's what a token is, how to estimate your costs, and why a new model can count the same text as more tokens.
By ScoviAI Team
Key points
- A token is a small chunk of text, often a word or part of one
- AI APIs charge per million tokens, with separate prices for input and output
- Each model family splits text its own way, so token counts differ
- Claude Haiku 5.5 counts the same text as about 30% more tokens than Haiku 4.5
If you've looked at AI pricing pages, you've seen prices like "$2 per million input tokens." But what is a token, and how many do you actually use?
What is a token?
AI models don't read whole words. They break text into small pieces called tokens. A common word like "the" is usually one token. A longer or rarer word may be split into two or three. Spaces, punctuation and numbers count too.
The tool that does the splitting is called a tokenizer, and every model family has its own.
How AI APIs charge for tokens
Developers who use AI through an API pay per token, usually quoted per million:
- Input tokens: everything you send, including your question, instructions and any documents.
- Output tokens: everything the model writes back. These usually cost several times more.
Here are a few current examples from this month's news:
| Model | Input (per 1M tokens) | Output (per 1M tokens) |
|---|---|---|
| Claude Haiku 5.5, prompts up to 100K | $0.10 | $0.50 |
| Gemini 4 Argon, introductory | $2 | $10 |
| OpenAI Decisions API | $0.10 | No charge |
Why the same text can cost more
Because every tokenizer splits text differently, the same paragraph can be a different number of tokens on different models. Anthropic says its new Claude Haiku 5.5 produces about 30% more tokens than Haiku 4.5 for the same text.
That doesn't automatically mean you pay more. Haiku 5.5's price per token is much lower, so most users still save. But it shows why you should compare the cost of your real task, not just the price per token.
How to estimate your costs
- Take a typical request, with your instructions, the user's message and any documents.
- Count its tokens with the provider's own token counter or a test run.
- Multiply input and output tokens by their prices.
- Multiply again by how many requests you expect each day.
Does this matter if I use ChatGPT or Claude apps?
Mostly not. Subscriptions like ChatGPT Plus or Claude Pro charge a flat monthly fee. Tokens still decide how long a conversation can get and when you hit usage limits, but you won't see a per-token bill.
Sources
- Introducing Claude Haiku 5.5 (Anthropic)
- Claude pricing (Claude Platform Docs)
ScoviAI Team is the ScoviAI newsroom. We write every article ourselves from the original announcements and link our sources. Spotted a mistake? Tell us and we'll fix it.
ADVERTISEMENT
