Context Window

The context window is the maximum amount of text (measured in tokens) the model can consider in a single call — system prompt, conversation history, retrieved docs, and the user's message all count toward it.

Frontier models in 2026 ship with 1M+ token windows (Claude Opus, Gemini, GPT-5). Bigger isn't always better: cost scales with input tokens and quality degrades on irrelevant context. The skill is choosing what to put in the window — relevant retrieval beats stuffing the whole knowledge base.

Example

An AI Engineer working on a bug pulls the relevant 4 files (~30K tokens) plus the issue description into context — not the whole repo (~2M tokens) which would cost more and answer worse.

How OpenLabor uses it

OpenLabor manages context per mission — retrieving what matters and dropping what doesn't.

Should I always use the largest context window I can?

No. Larger windows cost more and can dilute attention. Use what's needed, not what's available.

Related: llm, rag, memory.

AI Labor Glossary