Token (AI/LLM)

What Is Token (AI/LLM)?

LLMs do not read words — they read tokens. A token is a piece of text: common words are single tokens ("hello" = 1 token), while uncommon words are split into multiple tokens ("authentication" = 3 tokens). On average, 1 token ≈ 0.75 English words, or about 4 characters.

Why tokens matter: LLM pricing is per-token (input tokens are cheaper than output tokens), context windows limit how much text you can send (GPT-4o: 128K tokens, Claude 3.5: 200K tokens), and longer prompts = higher latency + cost.

Cost optimization: use shorter prompts, cache common responses, choose cheaper models for simple tasks, batch requests, and monitor token usage per feature.