What Are Tokens in AI? Explained With Actual Numbers

Stories you can play right now

‹

The Girl in the BoxA girl in a box.I Can Hear What They Are ThinkingThe only guy in an all-women dorm. Four minutes after moving in, he starts hearing what they think.Another World Under Twin Moons You've been teleported to a fantasy world where five princesses want to claim you in order to inherit the throne. My Tomboy Friend Turned Into a Mermaid?!{{user}}'s best friend has turned into a mermaid, and now they both have to figure out what to do about it...Chaotic Isekai with my Two Shameless BestiesThree childhood friends die together and awaken in a fantasy world, where two shameless otaku and a player chosen adventurer begin again.The Fox and the MarchEnding the war required a marriage and you're the lucky person!Crash Into Another WorldA routine drive ends in a crash—and you wake up in a world of magic, monsters, and deadly adventurers. Survive first; answers can wait.Last Call at the Sovereign A mysterious warrior hides as a bartender, serving four extraordinary women whose growing affections threaten {{user}}'s carefully guarded peaceful life.

›

Every AI app you use — chatbots, roleplay platforms, writing tools — bills, thinks, and forgets in tokens. Understanding them takes five minutes and permanently demystifies AI pricing and AI amnesia. Here's the plain-language version.

The short answer

A token is the chunk of text an AI model actually reads and writes. It's not quite a word: common words are one token, longer words split into pieces, punctuation counts. Rules of thumb worth memorizing:

• 1 token ≈ ¾ of an English word (100 tokens ≈ 75 words)
• A typical chat message: 30–80 tokens
• A long roleplay reply: 200–400 tokens
• This article: ~1,100 tokens

Non-English text is pricier: Japanese or Chinese can use 2–3× the tokens for the same meaning — one reason multilingual AI apps have uneven costs.

Why tokens decide what AI costs

Model providers charge per token, in and out — every message you send plus the conversation history the model re-reads plus its reply. That's why long conversations cost more per message than short ones: the model rereads the scene each turn. Apps package this differently: subscriptions (you subsidize heavy users or they subsidize you), energy meters (rationed tokens in disguise), or direct pay-per-use. We use the direct kind on ISEKAI ZERO — each model shows its per-message price in Arcane credits (100 = $1), so the token economics are visible instead of hidden.

Why tokens decide what AI remembers

A model's context window is measured in tokens — how much text it can consider at once. When your roleplay exceeds the window, the oldest lines fall out and the character "forgets." This is the entire technical explanation for AI amnesia. Fixes, in ascending sophistication: summarizing old scenes, lorebooks that re-inject key facts, and structured story memory that persists outside the window (our approach — plot state and character knowledge survive even when raw chat history doesn't).

Practical takeaways

• Long, rambling freeform chats burn tokens and memory quality; scene structure is cheaper and more coherent.
• "Unlimited" AI plans mean someone is rationing tokens somewhere you can't see.
• If a character forgot something crucial, it likely slid out of the context window — restate it or use a platform that persists it.

Want to see per-message token pricing in the wild instead of the abstract? Any storyline on ISEKAI ZERO shows each model's exact cost before you pick it.

Try ISEKAI ZERO free