What is a Token?
A Token is the smallest unit an LLM uses to process text — it could be a single character, a sub-word fragment, a common word, or punctuation. Models bill, count length, and measure speed in tokens, not in characters or words. Anthropic, OpenAI, and Google API pricing is all per-token.
Rough conversions:
- English: 1 token ≈ 0.75 words
- Chinese: 1 character ≈ 1-2 tokens (varies by model)
- Code: typically 5-15 tokens per line
Why it matters: estimating API costs, context window usage, and response speed all require thinking in tokens. Claude Opus 4.x’s 1M context window fits roughly 750K English words or 500K Chinese characters. A typical 3000-character Chinese blog post on Judy AI Lab is around 4500 tokens; the entire 75-term glossary is roughly 80K tokens. These numbers decide how you design your RAG system and prompt structure.