Tokens & Context Windows

How AI processes and "remembers" text.

Tokens:
- AI breaks text into chunks called "tokens"
- Roughly 4 characters per token (in English)
- "Hello world" ≈ 2 tokens
- Pricing is often per-token

Context window:
- How much the AI can "see" at once
- GPT-4: ~128K tokens
- Claude: ~200K tokens
- Gemini: ~1M tokens (announced)

Why it matters:
- Long documents may need to be chunked
- Earlier conversation may be "forgotten"
- More context ≠ always better results
- Large contexts cost more to process

Practical tip: If the AI seems to forget earlier instructions, the conversation may have exceeded its effective context.