Paste a prompt. The tool estimates token count using a heuristic and shows usage as a gauge against each model's context window. Heuristic only, actual tokens depend on each model's tokenizer (BPE for Claude, tiktoken for OpenAI, SentencePiece for Gemini, etc.). Useful for budget planning, not for billing-precise counts.
Notes on the estimate
- Heuristic: ~4 chars per token for English prose, weighted up for whitespace and punctuation. Real tokenizers vary.
- Code is denser: source code typically uses 3.0-3.5 chars per token. Long identifiers and operators tokenize differently from prose.
- Non-English: CJK and many other languages use significantly more tokens per character. The estimate undercounts here.
- Output budget: the context window is shared between input and output. If you're asking for a 20K-token response, your input budget is correspondingly smaller.
- Vision / files: images and document attachments consume tokens too. This tool only measures text.