For beginners

Get great answers, spend the fewest tokens

A token is a tiny piece of text — roughly a word or a few letters. You pay for every token the AI reads and writes. This guide shows how to say more with less, using the easiest possible examples.

The basics

First, what is a token?

You don't need to be technical to save tokens. Just understand this one idea.

A token = a small piece of text

Think of it like a coin at an arcade. Every word (or part of a word) you send costs a coin, and every word the AI sends back costs a coin too.

Your whole chat is re-counted each time

When you send a follow-up message, the AI re-reads the entire conversation so far — so old messages cost coins again and again.

Fewer tokens = less money

Shorter, clearer messages and shorter replies mean fewer coins spent. That's the whole game.

Six easy habits

How to use fewer tokens

Each habit below has a plain example you can copy.

1. Be specific and short

A tight instruction beats a long, chatty one. Tell the AI exactly what you want and the format.

Uses more tokens

Can you please, when you have a moment, write me something about dogs? I'm not sure what, maybe a few thoughts, whatever you think is nice. Thanks!

Uses fewer tokens

Write 3 fun facts about dogs in 2 sentences.

2. Trim the old conversation

Long chats get expensive because the AI re-reads everything. Start a fresh chat for a new topic, or delete turns you no longer need.

Uses more tokens

[200-line chat from this morning about taxes] ...and now help me write a birthday text to my mom.

Uses fewer tokens

New chat: "Write a warm birthday text to my mom, about 30 words."

3. Use the smallest model that does the job

Big, powerful models cost more per token. Use the smallest suitable model for easy tasks (summaries, lists, drafts), and save the big one for hard problems.

Uses more tokens

Use the biggest model to "turn this list into bullet points."

Uses fewer tokens

Use a smaller model to "turn this list into bullet points."

4. Cap the length of the answer

Tell the AI how long to keep it. In the API, set max_tokens. In chat, just say "keep it short."

{
  "model": "kimi-k3",
  "max_tokens": 200,
  "messages": [
    { "role": "user", "content": "Summarize this article." }
  ]
}

5. Put the standing rules in the system message

If you always want the same style or format, say it once in the system message instead of repeating it in every message.

Uses more tokens

Msg 1: "Be casual. Reply in bullets." Msg 2: "Be casual. Reply in bullets. Now fix my email." Msg 3: "Be casual. Reply in bullets. Now write a tweet."

Uses fewer tokens

System: "Be casual. Reply in bullets." Msg 1: "Fix my email." Msg 2: "Write a tweet."

6. Ask for the shape you want

Request the exact format — "3 bullets", "one sentence", "a 50-word summary" — so you don't get a long reply you'll trim anyway.

Uses more tokens

"Tell me about the product launch."

Uses fewer tokens

"Give me 3 bullet points about the product launch, under 40 words."

Before you hit send — quick checklist

Run through these in 5 seconds.

  • Did I remove chatty filler?
  • Did I start a fresh chat for a new topic?
  • Am I using the smallest model that fits?
  • Did I cap the answer length?
  • Are my standing rules in the system message?
  • Did I ask for the exact format?