"The bigger model performs well with a context window of up to a quarter million tokens, allowing you to analyze entire code bases, and/or multi-turn agentic use cases."