On-device, after download
MiniLM turns each sentence into 384 numbers. We compare the vectors using cosine similarity. The model runs in a Web Worker using CPU/WASM, without a server inference API.
A SMALL AI TOOL FOR A SPECIFIC TASK
Different words.
Similar meaning?
Compare a sentence with a few alternatives.
See which meaning comes closest.
Ready when you are. The model loads only when you click Compare meaning.
03 / CLOSEST MEANING FIRST
Cosine similarity from −1 to 1. A model score, not a probability or a plagiarism verdict.
KNOW WHAT YOU'RE MEASURING
A useful signal.
Keep your judgment.
This is a neural embedding model, not a keyword count. It can recognize similar wording that shares few words, and it can also miss important differences.
Your words stay here
We send no tool text to an AI service. Download providers receive ordinary requests for library and model files. Your browser may cache those files. No account, private API key or ads. Usage measurement is optional.
Short English text
The model uses at most 256 wordpieces per input; longer text is truncated. It is not a reliable judge of negation, factual truth, translation or Chinese text. Use it for comparing phrasing, not consequential decisions.
Free doesn't mean instant
Initial download is about 46MB; weak connections can take several minutes. Browser cache may be cleared. If a CDN is blocked, loading fails visibly. There is no hidden paid fallback.
Try a small experiment.
“A dog is running.” should be closer to “A puppy is running.” than to “The stock market fell.” Identical text should score near 1. Then try your own examples and inspect the failures.
How to read a similarity score ↗