Tokens: why some inputs cost more

A token is a chunk of text — often a word piece. English is roughly 1–2 tokens per word; code and other languages vary. You pay for input and output. A 20-token answer that sat on a 40,000-token paste still billed the paste.

• Repeating the same 30-page spec every turn is the classic own-goal.
• Images and video are not 'a few tokens.' They are their own meter.
• Tool loops resend the prefix. Ten rounds is ten prefixes.

Practical habit: log usage. Trim. Cap max_tokens. If you cannot explain last week's bill, you are not in production — you are in a fog.