# embeddings-benchmark/mteb

MTEB: State-of-the-art evaluation of embeddings across languages and modalities

Repository: https://github.com/embeddings-benchmark/mteb
Canonical: https://ross.abutalabs.com/products/mteb
Homepage: https://docs.mteb.org
Language: Python
License: Apache-2.0
License Family: permissive
Topics: benchmark, clustering, information-retrieval, sentence-transformers, sts, text-embedding, retrieval, neural-search, semantic-search, sbert, text-classification, bitext-mining, reranking, multilingual-nlp, mteb, low-resource-nlp, multimodal, embeddings, evaluation
Last push: 2026-08-26T21:33:26+00:00

## Health v2 (maintenance only)
Score: 95/100 (v2, computed 2026-09-03T02:39:23.370411+00:00)
- activity 99, release rhythm 87, longevity 100
- inputs: {"age_days": 1611, "days_push": 7, "days_rel": 8, "gap_med": 0, "n_releases_24m": 522}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 3406, forks 673 (observed 2026-08-28T04:08:03.094226+00:00)

## What it is
MTEB (Massive Text Embedding Benchmark) is a Python library and CLI for benchmarking embedding models across 1000+ tasks, languages, and modalities including text, image, and audio. It provides a standard evaluation harness, task selection, and a public leaderboard for comparing embedding and retrieval systems.

## Use cases
- evaluate embedding model quality on standard benchmarks
- compare sentence embedding models on retrieval tasks
- benchmark text embeddings across multiple languages
- test multimodal image-text embedding models
- run embedding evaluation from the command line
- generate model card metadata from benchmark results
- find the best embedding model for classification or reranking

## When to choose
- you need standardized, reproducible evaluation of embedding or retrieval models
- you want to compare your model against the MTEB leaderboard
- you need multilingual or multimodal (image/audio) embedding benchmarks
- you want a CLI or Python API for batch-running benchmark tasks

## When to avoid
- you need to train or fine-tune embedding models rather than evaluate them
- you need a production embedding inference service
- you only need simple cosine similarity utilities without benchmarking

## Facets
- artifact type: library
- maturity: active
- function: benchmarking, machine-learning, nlp, search-engine, cli
- domain: machine-learning, data-science
- platform: python, cli, cross-platform
- tags: embeddings, text-embedding, semantic-search, information-retrieval, sentence-transformers, evaluation, leaderboard, multilingual, reranking, benchmark-suite, natural-language-processing, search, multimodal

## Member repositories
- embeddings-benchmark/mteb (main) score 95

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:08:03.094226+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T18:38:36.264778+00:00, confidence not recorded.
  - readme: https://github.com/embeddings-benchmark/mteb (fetched 2026-08-28T04:08:03.094226+00:00, sha e37a62d2258f)
  - homepage: https://docs.mteb.org (fetched 2026-08-29T09:32:48.649280+00:00, sha 028c96e9e4dc)
  - site_page: https://docs.mteb.org/installation (fetched 2026-08-29T09:32:48.658641+00:00, sha f3893b66ce04)
  - site_page: https://docs.mteb.org/get_started/usage/get_started (fetched 2026-08-29T09:32:48.660286+00:00, sha 7f469a4ec9c8)
  - site_page: https://docs.mteb.org/get_started/usage/cli (fetched 2026-08-29T09:32:48.661879+00:00, sha 423d2fe904d4)
- Data as of 2026-08-30T08:39:29.467469+00:00.