Skip to content

Benchmark

MLEB - Contractual Clause 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
0.83

Score distribution

79%
Mean
80.1%
Median
9.7%
Std dev
71.1%
10th percentile
89.3%
90th percentile
51.3%
Min
51.3%92.8%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.83 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-largeContractual Clause Retrieval· MLEB data92.8%
2voyage-4Contractual Clause Retrieval· MLEB data91.5%
3voyage-4-liteContractual Clause Retrieval· MLEB data89.3%
4Qwen3 Embedding 4BContractual Clause Retrieval· MLEB data88.3%
5Qwen3 Embedding 8BContractual Clause Retrieval· MLEB data87%
6Text Embedding 3 LargeContractual Clause Retrieval· MLEB data86.8%
7snowflake-arctic-embed-l-v2.0Contractual Clause Retrieval· MLEB data81.1%
8Qwen3 Embedding 0.6BContractual Clause Retrieval· MLEB data80.9%
9Jina Embeddings v4Contractual Clause Retrieval· MLEB data80.1%
10Text Embedding 3 SmallContractual Clause Retrieval· MLEB data77.4%
11Gemini Embedding 001Contractual Clause Retrieval· MLEB data77%
12Jina Embeddings v5 Text SmallContractual Clause Retrieval· MLEB data76.3%
13Text Embedding Ada 002Contractual Clause Retrieval· MLEB data76.2%
14Jina Embeddings v5 Text NanoContractual Clause Retrieval· MLEB data73.1%
15BGE M3Contractual Clause Retrieval· MLEB data72.8%
16snowflake-arctic-embed-m-v2.0Contractual Clause Retrieval· MLEB data71.1%
17Multilingual E5 Large InstructContractual Clause Retrieval· MLEB data68.5%
18Granite Embedding Small English R2Contractual Clause Retrieval· MLEB data51.3%

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