MTEB
Benchmark suite ranking embedding models across many tasks.
Overview
MTEB evaluates embedding models across a wide set of tasks - retrieval, clustering, classification, reranking - in several languages, and publishes a single ranking. It solves the problem that "which embedding model" is otherwise answered by vendor claims on cherry-picked tasks. It exists because retrieval quality is decided at the embedding step, so a comparable ranking across models is the most useful single number in a RAG stack.
Technology Graph
How MTEB connects to the rest of the stack.
Uses and depends on
- BEIR benchmark Benchmarks MTEB based on BEIR High ✓ source
Applied to
- Embedding concept Techniques MTEB used for Embedding High ✓ source
Tools
Software that implements or integrates with MTEB.