AI Token Counter
Exact token counts for GPT models — the real tokenizer runs in your browser — plus labeled estimates for Claude and Gemini, context-window checks and prompt cost. Nothing is uploaded.
Exact where possible, honest where not
GPT counts here are exact: the same o200k_base / cl100k_base tokenizer the models use runs locally in your browser (first count may take a moment while it loads). Anthropic and Google don't publish their tokenizers, so Claude and Gemini counts are labeled estimates derived from the GPT count with calibrated ratios — close for planning, not billing-grade.
Why token counts matter
Tokens set both your cost and your context-window budget. A prompt that's fine in one model can overflow a smaller window in another; batch jobs priced per 1M tokens can differ 10x between model tiers. Counting before you send is the cheapest optimization there is. Writing the prompt first? Draft it in the word counter, then check tokens here.