MTEB
How the test works
Task
10 retrieval datasets in MTEB English v2
- What the model doesEmbed queries and documents, then rank relevant documents for each query
- Kinds of tasksText retrieval
- In → outText → Embeddings
Run
MTEB retrieval evaluation
Score
Mean retrieval nDCG@10 × 100
- Graded byCompare ranked documents with dataset relevance labels
- Better isHigher
MTEB English · Retrieval leaderboardtop 10 of 212
Full leaderboard with filters →| Model | MTEB English · Retrieval | Dimensions | Max input | Languages | Price | Released |
|---|---|---|---|---|---|---|
70.7unstated | 1,024 | 33K tokens | en | Nov 2025 | ||
70unstated | — | 33K tokens | — | May 2026 | ||
69.4unstated | 4,096 | 33K tokens | — | $0.01 / 1M tokens | Jun 2025 | |
68.8unstated | — | 33K tokens | — | Apr 2026 | ||
68.5unstated | 2,560 | 33K tokens | — | $0.02 / 1M tokens | Jun 2025 | |
67.5unstated | — | 33K tokens | — | Apr 2025 | ||
67.1unstated | 3,584 | 33K tokens | en, zh | Aug 2025 | ||
66.2unstated | 1,024 | 33K tokens | en, zh | Nov 2025 | ||
66.2unstated | 4,096 | 33K tokens | en | Jun 2025 | ||
64.4unstated | 3,072 | 2K tokens | — | $0.15 / 1M tokens | Jul 2025 |
Versions and boards on Model Menu
| Board | Models |
|---|---|
MTEB English · RetrievalShown above | 212 |
| 179 | |
| 89 | |
| 87 | |
| 37 | |
| 251 | |
| 238 | |
| 235 |
MTEB questions
What is MTEB?›
Embed queries and documents, then rank relevant documents for each query. It has 10 retrieval datasets in MTEB English v2. Created by MTEB community (originally Hugging Face and Cohere).
Which model is best on MTEB?›
Yuan Embedding 2.0 En leads MTEB English · Retrieval with 70.7, ahead of Ingot 8B R3 (70) and Qwen 3 Embedding 8B (69.4).
What is the best open-weights model on MTEB?›
Yuan Embedding 2.0 En with 70.7, then Qwen 3 Embedding 8B (69.4).
How is MTEB scored?›
Mean retrieval nDCG@10 × 100; compare ranked documents with dataset relevance labels.
Which versions of MTEB are there?›
MTEB English · Retrieval (212 models), MTEB multilingual · Reranking (179 models), MIEB lite · Document Understanding (89 models), MTEB multilingual (task avg) (87 models), MAEB v1 audio-only (task avg) (37 models), MTEB English · Pair Classification (251 models), MTEB English · Reranking (238 models), MTEB English · Classification (235 models).