Metric & Benchmark
aka Massive Text Embedding Benchmark
MTEB
Public benchmark and leaderboard scoring embedding models across 56 datasets.
Definition
MTEB unified embedding evaluation across retrieval, clustering, classification and similarity tasks. The HuggingFace leaderboard is the de facto reference for picking an embedding model; Voyage, OpenAI text-embedding-3, BGE and E5 routinely top it.
Common use cases
- Embedding selection
- RAG tuning
- Research benchmarks