T | TERSE | ISSUE 008 · AUGUST 2026 |
|
THE TERSE LETTER · 4 MIN READ Why “the” is 1 token
and “ChatGPT” is 2A 4-minute tour of tokenization — and the abbreviations that halve your prompts. |
You're not billed by the word or the character. You're billed by the token — and tokens don't map to either in the way you'd expect. Understand the rules and you can write prompts that say the same thing for far fewer tokens. |
Common words are cheap. Odd ones are expensive.Frequent words like "the" are a single token. Rare or invented words ("ChatGPT", "kubernetes", a long variable name) get split into pieces. Whitespace, punctuation, and capitalization all shift the count too. |
✗ MORE TOKENS repeated spaces ALLCAPS WORDS verbose synonyms snake_case_names |
| ✓ FEWER TOKENS single spaces normal case common words short names |
|
|
Abbreviations the model still understands."function" → "fn", "repository" → "repo", "configuration" → "config". The model reads them perfectly. Terse's aggressive mode applies a whole dictionary of these automatically — your meaning intact, your token count cut. |
2,000-word technical prompt | → On daily technical prompts, about $70/year. |
|
|
Aggressive mode applies safe abbreviations the model understands — automatically. |
Three things to do right now. |
1 | Normalize whitespace Collapse double spaces and stray line breaks. Terse soft mode does this for free. |
2 | Use abbreviations the model knows fn, repo, config, env, db — all instantly understood, all cheaper. |
3 | Turn on aggressive mode for drafts For throwaway prompts where polish doesn't matter, let Terse compress hard. |
|
macOS · Free forever tier · 1,500 optimizations/week |
|
NEXT WEEK Subagents and context hygiene. Spawning subagents is the cleanest way to keep your main thread cheap. I'll show the pattern I use every day. |
|
Until then — stay terse. — James
Builder of Terse · terseai.org |
