Context window

The context window of an LLM is the maximum number of tokens (roughly, fragments of words) the model can take as input in a single request. It determines how much information you can give the model to work with — the conversation history, retrieved documents, system instructions, examples, and so on.

2026 context windows for major models:

• Claude Sonnet 4.6: 1,000,000 tokens (roughly 750,000 words)
• Claude Opus 4.7: 200,000 tokens
• Claude Haiku 4.5: 200,000 tokens
• GPT-4o / GPT-5: 128,000 tokens
• Gemini 2 Pro: 1-2M tokens

Why context window matters for automation:

• RAG depth. Larger context windows let you retrieve more grounding documents per query. With 1M tokens, you can include entire knowledge bases.
• Multi-step reasoning. Agents executing long loops can carry full state across many steps.
• Long documents. Process whole contracts, codebases, or transcripts in one pass.

Trade-off: bigger context = more cost per call. A 100K token context call is meaningfully more expensive than a 5K token call. Production systems use techniques (RAG, summarization, sliding-window memory) to manage context size to what's actually needed for each call.