Sentence similarity checker
Compare sentences by meaning with a multilingual embedding model in your browser and see a cosine-similarity matrix.
Input
Result
The result will appear here.Embeddings are how RAG systems, deduplication and semantic search decide that two texts mean the same thing, and it helps to see the numbers before building on them. Enter sentences one per line and each is embedded by a small sentence-transformers model through huggingface/transformers.js, then every pair is compared by cosine similarity. The default multilingual model places Chinese and English paraphrases close together, so you can check whether a question and its translation really land near each other.
How it works
- Sentences are embedded with mean pooling and L2 normalisation, the way these sentence-transformers models were trained to be used.
- The output lists the sentences, a full similarity matrix and the ten most similar pairs, highest first.
- The multilingual model is about 118 MB and the English-only all-MiniLM-L6-v2 about 23 MB; both download from Hugging Face on the first run, and there is no progress bar while that happens.
Where your data goes
Nowhere. This tool runs entirely in your browser: the text you paste is processed by the page and is never transmitted to a server or written to a log.
This tool is free and needs no account. Its results exist only in your open page and are not saved anywhere.
What it costs
This tool is free, with no sign-in and no points.
Common questions
- What counts as similar?
- There is no universal threshold. With these models, paraphrases usually score well above unrelated sentences, and topic-related but different questions fall in between. Calibrate on your own data: put a few pairs you know are duplicates and a few you know are not into the box, and see where the line falls.
- Why does the first run seem stuck?
- Text tools on this site cannot show download progress, and the first run has to fetch the model from huggingface.co before it can compare anything. The multilingual model takes a while on a slow connection and may not load where Hugging Face is blocked. Later runs use the cached model and return in a second or two.
- Are these the same embeddings as OpenAI's or Qwen's API?
- No. Each embedding model has its own vector space, and scores from one are not comparable with another. This tool is for exploring how semantic similarity behaves and for small checks; for production retrieval, measure with the embedding model your system will actually use.
The open-source behind it
This tool runs on huggingface/transformers.js, released under Apache-2.0. If you need the same behaviour inside your own program, that is the library to reach for.
huggingface/transformers.jsAlso known as
- sentence similarity
- semantic similarity online
- cosine similarity calculator
- text embedding online
- compare two sentences