Sloane explains tokens
What is a token? AI limits, explained simply
Every AI plan talks about tokens, context windows and 5-hour limits. Sloane reads the fine print so you don't. Here's what it all means, in plain English and with pictures.
A token is a little chunk of text
AI doesn't read letter by letter or word by word. It reads in tokens: short words, pieces of longer words, numbers and punctuation.
A handy rule of thumb: 1 token ≈ ¾ of a word. So 100 tokens is about 75 words, and a page of text (~500 words) is roughly 650–700 tokens.
“I love pineapples!” → 3 words, but about 5 tokens
✍️ Try it: watch your text turn into tokens
⚠️ Approximate. This is a simple tokenizer that runs in your browser to show the idea. Real models (ChatGPT, Claude, Gemini…) each use their own tokenizer, so their counts will be a bit different.
The context window is the AI's desk
Picture the AI working at a desk. Everything it can see right now sits on that desk: your messages, its replies, any files you upload. That's the context window.
A bigger desk means it can keep more in view at once. When the desk fills up, the oldest papers slide off the edge: the AI starts to “forget” the start of the chat, or you get asked to start a new one.
- 32K tokens ≈ 24,000 words ≈ 48 pages
- 128K tokens ≈ 96,000 words ≈ 190 pages
- 1M tokens ≈ 750,000 words ≈ 1,500 pages
Approximate, using ~0.75 words per token and ~500 words per page.
Why long chats hit limits faster
Each time you send a message, the AI re-reads the whole conversation so far, not just your newest line. Message 20 costs far more than message 1, because it carries messages 1–19 with it.
Uploading big files and turning on reasoning (“thinking”) mode adds even more. Tip: start a fresh chat when you switch topics, and your allowance lasts longer.
If each message is ~200 tokens, the AI has read about 4,200 tokens in total so far, not just 1,200. (Illustration only.)
Each bar = what the AI re-reads to answer that message
How 5-hour and weekly windows work
Many plans (Claude, Gemini) give you a short window that refills about every 5 hours, plus a weekly cap on top. Think of a phone data plan with both a daily and a monthly limit.
Exact sizes and reset times differ by provider and plan. Most don't publish them; check the plan's page on Sloane.ai.
What “compute-based limits” mean
Instead of “40 messages”, more and more companies count how much computer work your requests take. Think of a fuel tank: a quick question uses a sip, while a long document, an image, a video or a deep-thinking answer uses a big gulp.
That's why there's often no fixed message number on the pricing page. It depends on what you ask for.
- 💬 Quick question
- 📄 Long document
- 🖼️ Make an image
- 🧠 Deep thinking / research
- 🎬 Make a video
Illustration only: real costs vary by provider
Now compare real plans
See the limits for every ChatGPT, Claude, Gemini, Grok, Perplexity, Copilot and Mistral plan in one table.
Open the comparison → Take the 1-minute plan picker →FAQ
How many words is a token?
About ¾ of an English word on average, so 100 tokens ≈ 75 words. Other languages, code and numbers often use more tokens per word.
How many pages is 128K tokens?
About 96,000 words, or roughly 190 pages of text (approximate, at ~500 words per page).
What is a context window?
How much text the AI can keep in mind at once: your messages, its replies and any files. It's measured in tokens.
Why do long chats hit limits faster?
Every new message makes the AI re-read the whole conversation, so later messages cost more of your allowance than early ones.