Token counter — how big is my text?

Characters, words, lines and estimated tokens for any text — prompt drafts, documents, code. Runs locally; nothing is uploaded.

Why token counts differ between tools

Tokens are the model's own vocabulary units, not words or characters. GPT-4o, Claude and Gemini each use different tokenizers, so the same text counts differently — sometimes by ±20%. Two practical rules:

• For budgeting, 4 chars/token is a solid planning heuristic.
• For exact counts (e.g. near a hard limit), use the provider's own tokenizer tool — and leave 10% headroom anyway.

What counts toward your context window?

• Your prompt (system + user + any pasted context)
• Previous messages in the conversation
• The model's reply (it also consumes context as it generates)
• Tool outputs and retrieved documents (RAG)

Need to shrink the text?

Use the prompt token reducer to cut filler and repetition, or the context optimizer for logs and JSON.