MTEB

MTEB, short for Massive Text Embedding Benchmark, is the standard benchmark for evaluating embedding models, introduced by Hugging Face and Cohere researchers in 2023. The benchmark covers 8 task types (classification, clustering, pair classification, reranking, retrieval, STS, summarization, bitext mining) across 56 datasets and 112+ languages. Each embedding model is evaluated on every task type and given an average score, producing the leaderboard at huggingface.co/spaces/mteb/leaderboard. Top performers as of 2025 include NV-Embed-v2 (NVIDIA), text-embedding-3-large (OpenAI), BGE M3 (BAAI), GTE-Qwen2-7B-instruct (Alibaba), and Voyage AI embeddings. MTEB has driven rapid improvement in the open-source embedding model ecosystem by providing a clear comparison metric. Variants include MTEB-FR (French), MIRACL (multilingual retrieval), CMTEB (Chinese), and MTEB-Vision for multimodal models. AI governance teams use MTEB as the primary benchmark when selecting embedding models for RAG deployments, though task-specific validation against domain data remains essential because MTEB averages can hide weaknesses on specific use cases.

MTEB-validated embeddings with Centralpoint: Centralpoint routes to MTEB-validated embedding models from OpenAI, Cohere, Voyage AI, NVIDIA, and BAAI in a model-agnostic stack. Tokens are metered per skill, prompts stay local, and retrieval-augmented chatbots deploy through one line of JavaScript on any portal.


Related Keywords:
MTEB,,