EngineersofAI LogoCoursesBlogResearch LabAI LettersThe LabCode BankInteractive 3DKodrEarnestJobs
Skip to main content
Interactive 3D/Embedding Evaluation - MTEB Benchmark Leaderboard
MTEB Scores - All Tasks
Retrieval (BEIR) (NDCG@10)
BGE
54.3
OpenAI
56.3
Cohere
55
STS (Spearman)
BGE
83.1
OpenAI
81
Cohere
80.1
Classification (Accuracy)
BGE
76.5
OpenAI
74.8
Cohere
73.9
Clustering (V-Measure)
BGE
51.2
OpenAI
49
Cohere
47.5
Reranking (MAP)
BGE
59.8
OpenAI
59.2
Cohere
58.1
Bitext Mining (F1)
BGE
92.5
OpenAI
94.1
Cohere
93.2
Controls
Models
Task Category
All tasks
Retrieval (BEIR)
STS
Classification
Clustering
Reranking
Bitext Mining
Primary Metric
NDCG@10
MAP
Spearman
MTEB (Massive Text Embedding Benchmark): 58 datasets across 8 task types. NDCG@10 for retrieval, Spearman for STS. No single model dominates all tasks.

Embedding Evaluation - MTEB Benchmark Leaderboard - Interactive Visualization

MTEB (Massive Text Embedding Benchmark) covers 58 datasets across 8 task types: retrieval, STS, classification, clustering, reranking, pair classification, and bitext mining. NDCG@10 is the primary metric for retrieval (via BEIR), Spearman correlation for STS tasks. No single model dominates all tasks. BGE and E5 often lead on STS, while proprietary models like OpenAI and Cohere lead on retrieval.

  • MTEB covers 8 task types - a model that excels at retrieval may be mediocre at STS
  • NDCG@10 for retrieval: measures quality of top-10 retrieved documents
  • Spearman correlation for STS: measures rank correlation with human similarity judgments
  • BGE-large-en-v1.5 and E5-large-v2 are top open-source models for most tasks
  • Model size vs performance: larger models help on retrieval but plateau on classification

Part of the EngineersOfAI Interactive 3D - free interactive visualizations covering every major concept in machine learning and AI engineering. Hover any element for a plain-English explanation. No code required.