Context Window Sizes Across Models

Sizes

Part of: The Context Window

The first widely used chat models had an 8,000-token window, about a dozen pages. A few years later, some models advertise windows of a million tokens or more, a stack of novels. That is a hundred-fold jump, and it completely changes what a model can do in one request. But bigger is not automatically better, and knowing why is what separates picking a model from guessing at one. What it is Different models ship with very different window sizes , measured in tokens. You'll see common tiers: - Small (about 8k tokens): roughly 6,000 words, a short essay or a few files. - Medium (about 32k tokens): roughly 24,000 words, a long report. - Large (128k to 200k tokens): a whole book in one request. - Huge (1M+ tokens): an entire codebase or a shelf of documents at once. Each tier is just a different ceiling on the same desk from lesson 1. The desk simply got bigger. How it works The window size is fixed per model, you don't grow it, you choose it by picking the right model for the task. The deciding factors are how much text the job genuinely needs in view at once, and what that costs: The logic is simple: pick the smallest window that comfortably fits the job, because the bigger windows co

Challenge: Model Right-Sizer