Budget your tokens before you send them

Why did my prompt get cut off, or my bill run higher than expected, when I never checked the token count first?

Beginner to intermediateRead 6 minDoing: 20 to 30 minutes the first time, a couple of minutes once it's routine6 steps

▶ Start

What you'll have at the end

A token count for a real prompt of yours, checked against your model's limit before you send it

Before you start

You need

A real prompt you're about to send: the fixed instructions, whatever background text you're including, and the actual message, plus your model's published token ceiling to look up.

Not covered

This does not cover shrinking a prompt with automatic summarization or retrieval, or estimating the dollar cost of a call; it only covers whether a prompt fits before you send it.

Leans on

See what you're really about to send

Gather the exact text you're sending
Count it with your model's real tokenizer

Check it against the budget before you send it

Find your ceiling and reserve reply room
Add it up and check against the ceiling

Fix it and prove the check holds

Cut what this question doesn't need
Prove the check on a different prompt

Where to go next

Log the token count and dollar cost behind every API call you make

for tracking what a call actually cost once you've sent it, the after-the-fact half of what this recipe checks before you send

Checked 15 Aug 2026

Part of the Large Language Models (LLMs) cookbook