A context window is the maximum amount of text (measured in tokens) a model can consider at once — the prompt, conversation history, and any retrieved documents. Larger windows enable longer briefs and codebases in one shot, but cost and attention quality do not scale linearly.
Think of the context window as the model’s short-term desk space: once it is full, older tokens fall off or must be summarized.
Vendors race to advertise larger windows because buyers equate them with “it can read my whole repo.”
Latent tests and deal coverage look at effective use of long context — not just the sticker length on a launch blog post.
Not automatically. Models can “lose” facts in the middle of long prompts, and pricing often charges for every token you stuff in.
Consumer and API models commonly advertise tens or hundreds of thousands of tokens; marketing numbers deserve independent verification on real tasks.