Skip to content
Models.

MTEB (and MMTEB) what it measures, how it is scored and who leads it

Embeddings · MTEB community (Hugging Face and collaborators) · introduced Oct 13, 2022 · Mean task score

Leaderboard Paper

Scores from MTEB Multilingual v2, published Sep 22, 2026

RankModelScore
1harrier-oss-v1-27b (microsoft)Microsoft · as “microsoft/harrier-oss-v1-27b”74.3%
2KaLM-Embedding-Gemma3-12B-2511 (tencent)Tencent · as “tencent/KaLM-Embedding-Gemma3-12B-2511”72.3%
3llama-embed-nemotron-8b (nvidia)Nvidia · as “nvidia/llama-embed-nemotron-8b”69.5%
4Qwen3-Embedding-8B (Qwen)Alibaba · as “Qwen/Qwen3-Embedding-8B”70.6%
5gemini-embedding-001 (google)Google · as “google/gemini-embedding-001”68.4%
6Qwen3-Embedding-4B (Qwen)Alibaba · as “Qwen/Qwen3-Embedding-4B”69.5%
7Octen-Embedding-8B (Octen)Octen · as “Octen/Octen-Embedding-8B”67.8%
8F2LLM-v2-14B (codefuse-ai)Codefuse Ai · as “codefuse-ai/F2LLM-v2-14B”68.7%
9F2LLM-v2-8B (codefuse-ai)Codefuse Ai · as “codefuse-ai/F2LLM-v2-8B”68.1%
10harrier-oss-v1-0.6b (microsoft)Microsoft · as “microsoft/harrier-oss-v1-0.6b”69.0%
11Seed1.6-embedding-1215 (Bytedance)ByteDance · as “Bytedance/Seed1.6-embedding-1215”70.3%
12F2LLM-v2-4B (codefuse-ai)Codefuse Ai · as “codefuse-ai/F2LLM-v2-4B”67.1%
13Giga-Embeddings-instruct-10B-A1.8B-0826 (ai-sage)Ai Sage · as “ai-sage/Giga-Embeddings-instruct-10B-A1.8B-0826”65.6%
14jina-embeddings-v5-omni-small (jinaai)Jinaai · as “jinaai/jina-embeddings-v5-omni-small”67.0%
14jina-embeddings-v5-text-small (jinaai)Jinaai · as “jinaai/jina-embeddings-v5-text-small”67.0%
16F2LLM-v2-1.7B (codefuse-ai)Codefuse Ai · as “codefuse-ai/F2LLM-v2-1.7B”65.2%
17BidirLM-Omni-2.5B-Embedding (BidirLM)Bidirlm · as “BidirLM/BidirLM-Omni-2.5B-Embedding”63.5%
18Giga-Embeddings-instruct-3B-0826 (ai-sage)Ai Sage · as “ai-sage/Giga-Embeddings-instruct-3B-0826”63.9%
19harrier-oss-v1-270m (microsoft)Microsoft · as “microsoft/harrier-oss-v1-270m”66.5%
20Qwen3-Embedding-0.6B (Qwen)Alibaba · as “Qwen/Qwen3-Embedding-0.6B”64.3%
21jina-embeddings-v5-omni-nano (jinaai)Jinaai · as “jinaai/jina-embeddings-v5-omni-nano”65.5%
21jina-embeddings-v5-text-nano (jinaai)Jinaai · as “jinaai/jina-embeddings-v5-text-nano”65.5%
23gte-Qwen2-7B-instruct (Alibaba-NLP)Alibaba · as “Alibaba-NLP/gte-Qwen2-7B-instruct”62.5%
24BidirLM-1.7B-Embedding (BidirLM)Bidirlm · as “BidirLM/BidirLM-1.7B-Embedding”63.5%
25ICT-TIME-and-Querit-embedding-v1 (ICT-TIME-and-Querit)Ict Time And Querit · as “ICT-TIME-and-Querit/ICT-TIME-and-Querit-embedding-v1”63.8%

About this benchmark

Task
Embed text for retrieval, clustering, classification, reranking and similarity.
Dataset
v1: 8 task types, 58 datasets, 112 languages. MMTEB: over 500 tasks in 250+ languages.
Method
Each embedding model is run on every task with the task's own metric; the leaderboard averages them.
Metric
Mean task score
Organization
MTEB community (Hugging Face and collaborators)
Introduced
Oct 13, 2022
Paper
MTEB: Massive Text Embedding Benchmark (arXiv 2210.07316)
Official leaderboard
huggingface.co/spaces/mteb/leaderboard
Home
github.com/embeddings-benchmark/mteb

Versions

Newest first. New versions are added, never rewritten.

VersionDate
MMTEBMassive multilingual expansion (arXiv 2502.13595); our board reads MTEB Multilingual v2.Feb 19, 20251 year ago
1Released with the paper.Oct 13, 20223 years ago

Sources: each benchmark's paper (arXiv) and its official site or leaderboard. Scores are the boards captured here every day, exactly as published.

New #1s by email

Saturdays, only in weeks when a leaderboard has a new #1.

Double opt-in. Unsubscribe any time.