Context budget checker
A rough token estimate (about four characters per token for English, less for code) compared against the common window sizes. Good enough to know whether you are at 12% or 120%.
0estimated tokens
0characters
0words
0lines
- 8,000 0%
- 32,000 0%
- 128,000 0%
- 200,000 0%
- 1,000,000 0%
Why the estimate can be off
English prose averages about four characters per token; code, JSON, and anything with lots of punctuation runs closer to three, and non-Latin scripts can be much worse. If the number matters, measure it with the provider's tokenizer instead of trusting this.
What people get wrong
The window is shared: system prompt, tool definitions, conversation history, retrieved documents, and the answer all compete for it. A 128K window with a 20K system prompt and six tools is not a 128K window.