# Twenty messages. 91% of the month.

> The question was never the expensive part.

- **Format:** short video, 9 steps, ~51 seconds
- **Topic:** What an AI token is and how AI cost is calculated — a token is a subword fragment, Dart code measured about 6.5 tokens per line, and because every chat turn resends the whole conversation and every attached file, one twenty-turn thread can use most of a monthly allowance.
- **Author:** Suthahar Jegatheesan (MSDEVBUILD)
- **Category:** GitHub Copilot · copilot
- **Published:** 2026-04-12
- **Tags:** ai, llm, githubcopilot, tokens, aicost, promptengineering, developerproductivity, msdevbuild
- **Canonical URL:** https://blog.msdevbuild.com/shorts/ai-token-cost-every-turn-resends/

---
## What you'll learn

- What a token is, and why it is not a word
- What really gets counted in one Copilot chat turn
- Why a long thread costs far more than short ones

## Understand it one step at a time

### 1. A token is a fragment

Not a word, not a character. "Jegatheesan" is 3 tokens.

### 2. Your code, measured

Three real Flutter files: 5,989 tokens, about 6.5 per line.

### 3. What one turn carries

System prompt, tools, instructions, files, then your 50-token question.

### 4. Every turn resends it all

Turn 20 carries turn 1 again, plus nineteen questions and replies.

### 5. 91% in one thread

Twenty turns: about 351,800 input and 12,000 output tokens.

### 6. Output costs more

Input is one parallel pass. Output is generated one token at a time.

### 7. Short threads, small attachments

Five threads of four turns, attaching only the method: about a third of the month.

### 8. Measure your own code

Five lines of tiktoken give you your own tokens per line.

### 9. Every turn resends everything

A token is a fragment; the whole conversation and every file travel on every turn.

---

## The takeaway

**Every turn resends everything.**

Measure your tokens per line, keep threads short, attach only what you mean.