LLM Token Counter: See exactly how GPT models split your text.

A token is the unit a language model reads and is billed by: a common word, a piece of a longer word, a punctuation mark or a run of spaces. The tokenizer splits text into tokens from a fixed vocabulary learned during training, so frequent English words are usually one token each while rare words break into several. As a rough guide, English prose averages around four characters per token, which is why the counter shows characters per token next to the total.

They are the two OpenAI byte-pair encodings, named after their vocabulary sizes of roughly 200,000 and 100,000 tokens. As of October 2026, o200k_base is used by GPT-5, GPT-4.1, GPT-4o and the o-series reasoning models, and cl100k_base by GPT-4, GPT-3.5 Turbo and the text-embedding-3 models. The larger vocabulary usually needs fewer tokens for the same text, especially outside English, so pick the encoding that matches the model you are sending to.

Each model family ships its own tokenizer with its own vocabulary, so the same text splits into a different number of tokens. This counter implements only the OpenAI encodings, which makes it an approximation for any other model — useful for sizing a prompt, not for billing. For an exact figure, use the vendor's own token-counting endpoint or the usage numbers its API returns.

Byte-pair encodings work on UTF-8 bytes, and the vocabulary is dominated by text that was common in training data. An emoji or a CJK character takes three or four bytes, and when that byte sequence is not a single vocabulary entry it is split into several tokens. That is why a short line of emoji can cost more than a full English sentence, and why o200k_base, with its larger multilingual vocabulary, often counts CJK text lower than cl100k_base.

Type the price your provider charges per million tokens and the counter multiplies it by the token count to show an estimated cost. No prices are built in, because they change often and differ by model, tier and region. Input and output tokens are usually priced differently, so enter the input price to cost a prompt and the output price to cost a reply of the same length.

Yes. Code is full of punctuation, operators, brackets and indentation, and each of those tends to split into its own token or a short run, so code usually yields fewer characters per token than English prose. Runs of spaces are often merged into a single token, which helps indented code, but unusual identifiers and long string literals still break into several pieces. Paste a real file to see the split rather than estimating from its length.

The counter tokenizes exactly the text you paste. A chat completion request wraps every message in role and separator tokens, which adds a few tokens per message plus a few for the reply, and tools or a system prompt add their own. Expect the API's prompt usage to be slightly higher than this count for a single message, and higher still for a long conversation.

No. The token counter runs 100% client-side: the tokenizer is loaded from this site into the page and your text is tokenized in your browser tab. Nothing is uploaded to a server, and there is no account or signup, so you can safely paste internal prompts or documents.