Benchmark
MLEB - Legal RAG Bench
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
- 1.06
Score distribution
- 40.1%
- Mean
- 39.8%
- Median
- 10.5%
- Std dev
- 27%
- 10th percentile
- 51.2%
- 90th percentile
- 23.8%
- Min
This benchmark has a weight of 1.06 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
| # | Model | Score |
|---|---|---|
| 1 | voyage-4-largeLegal RAG Bench· MLEB data | 64.8% |
| 2 | voyage-4Legal RAG Bench· MLEB data | 56.6% |
| 3 | voyage-4-liteLegal RAG Bench· MLEB data | 51.2% |
| 4 | Qwen3 Embedding 8BLegal RAG Bench· MLEB data | 50% |
| 5 | Qwen3 Embedding 4BLegal RAG Bench· MLEB data | 45.4% |
| 6 | Jina Embeddings v5 Text SmallLegal RAG Bench· MLEB data | 42.3% |
| 7 | Gemini Embedding 001Legal RAG Bench· MLEB data | 42.2% |
| 8 | snowflake-arctic-embed-l-v2.0Legal RAG Bench· MLEB data | 40.2% |
| 9 | Text Embedding 3 LargeLegal RAG Bench· MLEB data | 39.8% |
| 10 | Qwen3 Embedding 0.6BLegal RAG Bench· MLEB data | 38.7% |
| 11 | Jina Embeddings v5 Text NanoLegal RAG Bench· MLEB data | 38.6% |
| 12 | Jina Embeddings v4Legal RAG Bench· MLEB data | 37.5% |
| 13 | Text Embedding 3 SmallLegal RAG Bench· MLEB data | 37% |
| 14 | snowflake-arctic-embed-m-v2.0Legal RAG Bench· MLEB data | 32.8% |
| 15 | Text Embedding Ada 002Legal RAG Bench· MLEB data | 27.4% |
| 16 | Multilingual E5 Large InstructLegal RAG Bench· MLEB data | 27% |
| 17 | BGE M3Legal RAG Bench· MLEB data | 26.8% |
| 18 | Granite Embedding Small English R2Legal RAG Bench· MLEB data | 23.8% |
Data credits: MLEB (Isaacus), MIT (licence)