About the Context Window Estimator
This tool shows what percentage of common LLM context window sizes — 8K, 32K, 128K, 200K, and 1M tokens — a piece of text or document would use. Paste text directly, or upload a plain text file (.txt, .md, .csv) or a PDF, and the tool extracts and counts it locally. Token counting uses a genuine BPE tokenizer implementation (the same algorithm behind OpenAI's tiktoken), giving an accurate, verifiable count rather than a rough guess. Since context windows are typically shared between your input and the model's expected response, an adjustable "reserve for response" setting is subtracted from the available space before calculating how full each window would be. All processing — including PDF text extraction — happens locally in your browser; your file is never uploaded anywhere.
Accuracy Notes
Real tokenizerCounts are computed with an actual BPE tokenizer implementation, not a character-count guess — the number reflects genuine tokenization behavior.
Model-specific varianceDifferent model families (GPT, Claude, Gemini, open-weight models) use different tokenizers, so the exact count for a specific model can differ somewhat from what's shown here — this is a solid general-purpose estimate, not a guaranteed exact figure for every provider.
Shared input/output spaceA context window isn't exclusively for your input — the model's response, and often the full conversation history, share the same budget. The reserve setting approximates this; real usage also depends on system prompts and prior messages you haven't included here.
PDF extractionText extraction only works for PDFs with actual selectable text. Scanned or image-only PDFs won't extract meaningful text without OCR, which this tool doesn't perform.
Window sizes shown8K, 32K, 128K, 200K, and 1M are common reference sizes across current LLMs — a specific model you're targeting may use a different exact figure.
Send Feedback
For any questions, bug reports, or feature requests, please drop your comments below.