Semantic Sentence Similarity Calculator

Encode two sentences with multilingual MiniLM and display their cosine similarity as a percentage with an interpretive range.

Focus mode

Sentence A

Sentence B

Result

Cosine similarity between sentence embeddings. Semantically similar sentences can score highly even when they use different languages.

Enter both sentences and select “Calculate similarity.”

Key features

Encodes two sentences as normalized multilingual MiniLM embeddings
Calculates cosine similarity between the embeddings as a percentage
Shows a very similar, similar, somewhat related, or low relation range
Uses the same comparison for sentences written in different languages

How to use

Enter the two sentences you want to compare.
Calculate similarity and review the percentage and interpretive range.

Examples

Compare sentence candidates
Use model-based semantic similarity to prioritize pairs for human review.
Reference cross-language wording
Inspect embedding similarity across languages, then verify the actual translation separately.

Privacy

Both sentences are processed inside your browser for MiniLM inference. After you consent on first use, model files may be downloaded from Hugging Face and cached in browser storage.

Requirements

The first comparison needs about 135 MB of model files. The score and range are model-based reference values and do not prove that two passages have identical meaning.

Common questions

Can I compare sentences in different languages?: Yes. The multilingual MiniLM model compares them through the same embedding process. The score is a model-based reference, not a judgment of translation accuracy.
Does a high percentage prove identical meaning?: No. It reports cosine similarity between normalized embeddings and does not guarantee agreement in facts or negation.