Tokenizer Mono

A monospace font that highlights LLM tokens as you type. Every highlight is drawn by the font itself: OpenType substitution rules find the token boundaries and a color table paints them. No JavaScript touches the text.

One typographic family, 28 faces, one per tokenizer: OpenAI's cl100k and o200k, Llama 2 through 4, Gemma, Mistral, Qwen, DeepSeek, Kimi K2, BERT, T5 and more. Each face agrees with its real tokenizer on 97 to 99% of token boundaries, measured against a mixed corpus of prose and code.

The same sentence in every face

@font-face {
  font-family: "Tokenizer Mono cl100k";
  src: url("TokenizerMono-cl100k.woff2") format("woff2");
}
textarea { font-family: "Tokenizer Mono cl100k", monospace; }

Two palettes are built in, for light and dark backgrounds, selected with the CSS font-palette property. Anywhere color fonts aren't supported, it degrades to plain IBM Plex Mono.

Install

Download tokenizer-mono-fonts.zip (TTF, WOFF, WOFF2 for every face) or tokenizer-mono-woff2.zip (web files only) from the latest release. Font Book on macOS registers the family with all 28 styles.

Read More