Comprehensive Guide to LLM Tokens

What exactly is a Token?

Large Language Models (LLMs) like ChatGPT, Claude, and Gemini don't read text letter-by-letter or word-by-word. Instead, they break text down into chunks called Tokens. A token can be an entire word, a syllable, a single character, or even a blank space.

As a general rule of thumb for English text: 1 token ≈ 4 characters or 100 tokens ≈ 75 words. However, formatting, code, non-English languages, and special symbols can drastically change this ratio.

Input vs. Output Tokens

When you interact with an API, you are billed for two distinct phases:

• Input Tokens (Prompt): The text you send to the AI. This includes your system instructions, previous conversation history, and the current query. This is usually significantly cheaper.
• Output Tokens (Completion): The text the AI generates in response. Because generating text requires significantly more compute power than reading it, output tokens are usually 2x to 4x more expensive than input tokens.

Tips to Optimize API Costs

• Trim the fat: Remove unnecessary whitespace, trailing spaces, and redundant instructions from system prompts.
• Minimize history: If building a chatbot, don't send the entire conversation history every time. Implement a sliding window or summarize older messages.
• Choose the right model: Don't use GPT-4o or Claude Opus for simple extraction or formatting tasks. Use smaller, significantly cheaper models like GPT-4o-mini or Claude Haiku.

Popular Model Pricing (per 1 Million Tokens)

Model	Input Price	Output Price