Lesson 438 of 1570
What a Token Actually Is (And Why It Matters for Your Prompts)
AI doesn't read words — it reads tokens. Knowing the difference makes you a better prompter.
Lesson map
What this lesson covers
Learning path
The main moves in order
- 1What a Token Actually Is (And Why It Matters for Your Prompts)
- 2What an AI Call Actually Costs (Tokens, Pricing, and Why Limits Exist)
- 3The big idea
- 4AI and Tokens: Why AI 'Forgets' Mid-Conversation
Concept cluster
Terms to connect while reading
Section 1
What a Token Actually Is (And Why It Matters for Your Prompts)
AI doesn't read words — it reads tokens. Knowing the difference makes you a better prompter.
What to actually do
- English averages ~0.75 words per token
- Numbers, code, and rare words use way more tokens
- Long prompts = real cost; concise prompts = real savings
Key terms in this lesson
The big idea: AI thinks in tokens, not words. Once you see the split, you write better prompts.
Section 2
What an AI Call Actually Costs (Tokens, Pricing, and Why Limits Exist)
Section 3
The big idea
Every AI call costs the company GPU compute, which they price per million tokens of input and output. As of 2026, GPT-5 costs roughly $5/million input tokens and $20/million output tokens; Claude Sonnet 4.5 is ~$3/$15. A typical paragraph response is ~300 output tokens — so a single ChatGPT response costs the company a fraction of a cent. This explains everything: why free tiers throttle, why heavy use of vision/voice gets cut off, and why building your own AI app is now within a teen's budget ($20 of API credit can run hundreds of experiments).
Some examples
- openai.com/api/pricing and anthropic.com/pricing list current per-token rates — bookmark these if you ever build with APIs.
- 1 million tokens ≈ 750,000 words ≈ a long novel — so $5 buys you 750K words of GPT-5 input.
- OpenAI's Playground gives free trial credits ($5 for new accounts as of 2024) — enough to learn the API for a weekend.
- If a free chatbot suddenly shows 'limit reached,' that's the company protecting its compute bill — usually resets in 3-5 hours.
Try it!
Open OpenAI's tokenizer (platform.openai.com/tokenizer). Paste any paragraph of yours and see exactly how many tokens it is. Now you can think in tokens, not words.
Section 4
AI and Tokens: Why AI 'Forgets' Mid-Conversation
Section 5
The big idea
AI reads in 'tokens' — chunks like words or word-pieces. Each model has a max it can hold (the 'context window'). When you exceed it, the oldest stuff drops off — that's why ChatGPT 'forgets' something from earlier. GPT-4o holds about 128k tokens (~96k words). Claude can hold a million. Knowing this means you stop blaming the AI and start managing context.
Some examples
- 1 token ≈ 0.75 English words.
- GPT-4o context window: ~128,000 tokens.
- Claude Opus: up to 1 million tokens.
- When AI 'forgets', re-paste the key info.
Try it!
Ask AI: 'How many tokens are you holding right now?' Some can answer. Then count words in your chat — divide by 0.75.
Key terms in this lesson
End-of-lesson quiz
Check what stuck
15 questions · Score saves to your progress.
Tutor
Curious about “What a Token Actually Is (And Why It Matters for Your Prompts)”?
Ask anything about this lesson. I’ll answer using just what you’re reading — short, friendly, grounded.
Progress saved locally in this browser. Sign in to sync across devices.
Related lessons
Keep going
Builders · 40 min
AI and tokens vs words: why your prompt costs what it costs
Learn what a token actually is so you can predict cost and context limits.
Builders · 40 min
Why AI 'Forgets' Halfway Through a Long Chat
AI has a memory limit called the context window. Hitting it explains a LOT of weird behavior.
Explorers · 40 min
Why AI Forgets the Start of a Long Chat
AI has a memory limit for how much of a chat it can remember at once.