Terminology· 2 min
What is a context window?
The context window is the maximum amount of text a model can consider at once — inputs and outputs combined.
Key idea
Measured in tokens (~4 characters each). Bigger windows let models read whole books, but cost more per call and can lose focus.
Example
GPT-4o Turbo: 128k tokens (~300 pages). Gemini 1.5 Pro: up to 2M. Claude Sonnet: 200k.
What to remember
- Long context ≠ perfect recall (see 'needle in a haystack' tests)
- Costs and latency grow with input length
- Retrieval (RAG) is often cheaper than dumping everything in