A "128K context window" doesn't mean 128,000 tokens of room for your document. Part of that budget is typically reserved for the model's own response, and knowing how much is actually left for your input is the difference between a document that fits and one that gets silently truncated.
How the Check Works
Given a token count and a context window size, the actual available space for input is:
Available for Input = Context Window Size − Reserved Tokens
"Reserved tokens" is space set aside for the model's expected response — if you're asking for a long, detailed answer, reserving more of the window for that response leaves proportionally less room for what you paste in.
Common Context Window Sizes
- 8K — smaller/older models, or budget-tier API configurations
- 32K — mid-range context, common for many production use cases
- 128K — large enough for most full documents, research papers, or codebases
- 200K — used by some of the largest current models
- 1M — the newest, largest context windows, capable of holding entire books
Why a Percentage Is More Useful Than a Raw Count
Knowing a document is "7,500 tokens" doesn't immediately tell you anything until it's compared against a specific window. That same document uses 93.8% of an 8K window — nearly full — but only 5.9% of a 128K window and 0.8% of a 1M window. Checking the percentage against each common window size at once makes it immediately clear which models can actually handle your input and which can't.
Check your own text against every common window size at once
Try the Context Window Estimator