Skip to content

Benchmark

MLEB - License TL;DR Retrieval

Embedding benchmarks measure how well models find related items, for example when matching a search to relevant results. Scores are percentages.

Models measured
18
Scores recorded
18
Sources
1
IQ weight
0.87

Score distribution

67.2%
Mean
66.7%
Median
6.8%
Std dev
59.9%
10th percentile
76.4%
90th percentile
54.1%
Min
54.1%81.4%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.87 for embedding models, measured over 18 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1voyage-4-largeLicense TL, DR Retrieval· MLEB data81.4%
2voyage-4License TL, DR Retrieval· MLEB data77.6%
3Qwen3 Embedding 4BLicense TL, DR Retrieval· MLEB data76.4%
4Qwen3 Embedding 8BLicense TL, DR Retrieval· MLEB data73.3%
5voyage-4-liteLicense TL, DR Retrieval· MLEB data71.8%
6Jina Embeddings v5 Text SmallLicense TL, DR Retrieval· MLEB data71%
7Gemini Embedding 001License TL, DR Retrieval· MLEB data69.1%
8Jina Embeddings v5 Text NanoLicense TL, DR Retrieval· MLEB data67.6%
9Text Embedding 3 LargeLicense TL, DR Retrieval· MLEB data66.7%
10Qwen3 Embedding 0.6BLicense TL, DR Retrieval· MLEB data64.4%
11snowflake-arctic-embed-m-v2.0License TL, DR Retrieval· MLEB data64.3%
12Jina Embeddings v4License TL, DR Retrieval· MLEB data64.1%
13Granite Embedding Small English R2License TL, DR Retrieval· MLEB data63.6%
14BGE M3License TL, DR Retrieval· MLEB data63%
15snowflake-arctic-embed-l-v2.0License TL, DR Retrieval· MLEB data62.4%
16Text Embedding 3 SmallLicense TL, DR Retrieval· MLEB data59.9%
17Text Embedding Ada 002License TL, DR Retrieval· MLEB data59.2%
18Multilingual E5 Large InstructLicense TL, DR Retrieval· MLEB data54.1%

Data credits: MLEB (Isaacus), MIT (licence)