Similar words by vector

Type a word or phrase and the tool shows the words and phrases with the nearest vector (embedding), the ones a language model sees as close in meaning. Any language works: “coffee” and “кава” land next to each other.

Try:

Free · no sign-up · up to 10 searches per hour · any language

Compare two phrases

How close two meanings are

What “similar by vector” means

A language model can turn any text into an embedding: a list of numbers, 1536 for text-embedding-3-small, that describes what the text means. Texts about the same thing get lists that point in nearly the same direction, even when they share no words or are in different languages: “coffee”, “espresso” and “кава” end up close together, “football” far away.

How close two vectors point is measured by cosine similarity, from 0 (unrelated) to 1 (the same meaning). Semantic search, recommendations and chatbots that answer from your documents all work this way. This tool lets you see it: which words the model puts next to yours, and how far apart they are.

How it works

The model turns text into a vector of 1536 numbers. Texts close in meaning get close vectors, so similarity can be measured instead of guessed.

  1. 01

    Candidates

    A language model suggests synonyms, associations, translations and a few deliberately unrelated words.

  2. 02

    Vectors

    Your phrase and all the candidates become embeddings in one request.

  3. 03

    Similarity

    Each candidate gets its cosine similarity to your phrase. The order comes from the numbers, not from the model’s opinion.

  4. 04

    Map

    The vectors are squeezed into two dimensions to show how the words group by meaning.

FAQ

Didn’t find the answer?

Write to me — I will reply within one business day.

What is an embedding?

A list of numbers a language model uses to describe what a text means. Texts about the same thing get close lists even when the words or languages differ. Semantic search, recommendations and chatbots that answer from a site’s documents all run on embeddings.

Why is the similarity rarely above 0.8?

With text-embedding-3-small even synonyms usually score 0.6–0.8, and unrelated words 0.1–0.25. What matters is the order, not the numbers: the higher a word is in the list, the closer it is in meaning.

Where do the words on the map come from?

A language model suggests them: synonyms, associations, translations and a few unrelated words for contrast. The similarity and the place on the map are computed from the vectors, so the model cannot bend the result.

How many searches can I run?

Up to 10 new searches per hour from one address. A phrase someone has searched before comes from the cache and does not count.

Need smart search on your site?

I build search by meaning, not by matching words: a catalogue, a knowledge base or a chatbot that answers from your site’s documents.

Assistant
Online now