Module 2.2 · Topic 2
Understanding Context Windows
Bottom Line Up Front: Each model has a context window: the max information it can process in one turn. Most context windows vary from 4K to 1M+ tokens. Matching window to task (size and content) creates better results.…
2.1 How Context Windows Work
Windows vary by task: 4K-16K is useful for most simple single-turn chats that don't require thinking, reasoning or research; 32K-100K is a helpful minimum range when working with documents or for longer conversations; 200K - 1M is a comprehensive range for multiple documents, large projects, very long conversations and for heavy research. Larger windows don't guarantee better quality, and focus is not necessarily consistent across all information sets within a very large context window. Match window to task.
2.2 Practical Limits of Context Length
One turn or multiple? As your conversation goes on, the context grows and the focus window shifts. Pulling in additional research and sources compounds the context accumulation. Imagine a conversation that starts by analyzing a single court judgment, and carries on by pulling in other resources to dig into topics found in the initial analysis? With a 100,000 token context window, it won't take long for parts of the source document or earlier steps in the conversation to fall outside the moving context window that feeds the next turn of your conversation.
Unlock the Full AI Skill Building Experience
Visit LawQi for access to the full AI Skill Building experience.
Visit LawQi40% discount using code REASONABLE for personal subscriptions.
Free 48-hour preview access when investigating for teams.