Token Counter by Model
Count tokens for any AI model in real time. Paste your prompt to see exact token counts for OpenAI models or fast estimates for Claude, Gemini, and more.
0
0
0
Frequently Asked Questions
How are tokens counted?
For OpenAI models (GPT-5.4, GPT-5.4 Mini, etc.) this tool uses gpt-tokenizer, a pure JavaScript implementation of OpenAI's byte-pair encoding (BPE) algorithm, running entirely in your browser. The count is exact. For other providers — Anthropic Claude, Google Gemini, Meta Llama — their tokenizers are not publicly distributed as client-side libraries, so we use a character-ratio estimate (≈ 1 token per 4 characters), which is accurate to within a few percent for typical English text.
Why do different models have different token counts?
Each AI provider trains their own tokenizer — a vocabulary of subword units that maps text to integers. OpenAI's o200k_base (used by GPT-5.4) has a larger vocabulary than older encodings, so it tends to encode the same text into fewer tokens. Claude and Gemini use proprietary SentencePiece-based vocabularies. The differences are usually small (< 10%) for English prose but can be significant for code or non-Latin scripts.
What tokenizer does this tool use?
We use the gpt-tokenizer npm package, a pure JavaScript BPE tokenizer for OpenAI models. It supports both the cl100k_base encoding (GPT-4.1 Legacy) and o200k_base encoding (GPT-5.4, GPT-5.4 Mini, GPT-5.4 Nano). All computation happens locally in your browser — no text is sent to any server.