MTEB: Massive Text Embedding Benchmark
MTEB just became the standard. 58 embedding models benchmarked on a single leaderboard—and your favorite model is probably losing.
Why it matters
MTEB established a unified evaluation framework for text embeddings across 56 datasets, enabling direct model comparison and driving competition in a critical but previously fragmented category. For builders choosing embedding models, this benchmark became the decision arbiter.
The key facts
6 to know56 datasets across retrieval, clustering, semantic similarity, and reranking tasks
58 embedding models benchmarked on unified leaderboard
Covers both open-source and proprietary models
Published October 2022 on Hugging Face
First standardized evaluation framework for text embeddings at scale
Enables direct comparison of embedding quality across use cases
Go to the source
Hugging Face Bloghuggingface.co