THE THESAURUS FOR THE LLM ERA · 0.9

See how AI
reads your words.

Look up any word and see how models actually split it into tokens, where it sits among their learned associations. Simple by default, with the raw numbers one tap away.

Also in 4 more languages, all measured against the English word they translate
Community
Facebook soon Discord soon

What you get per word

20,660 English words, plus 123,338 headwords across four more languages.

Measured

PanelSource
Definition & sensesWordNet
Token split per modelthe models' own tokenizers
Personality axesreal embedding matrices
Neighbours & surprisesreal embedding matrices

Extracted from Qwen3 8B, DeepSeek V3 and Mistral 7B, one embedding shard each, no GPU and no inference.