Skip to content

Benchmark

Agentset Embedding Recall@10 - MSMARCO

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

Models measured
15
Scores recorded
15
Sources
1
IQ weight
0.89

Score distribution

21.9%
Mean
22.1%
Median
0.429%
Std dev
21.2%
10th percentile
22.3%
90th percentile
20.8%
Min
20.8%22.4%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.89 for embedding models, measured over 15 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1Qwen3 Embedding 4BMSMARCO· Agentset22.4%
2Qwen3 Embedding 8BMSMARCO· Agentset22.3%
3Text Embedding 3 LargeMSMARCO· Agentset22.3%
4voyage-3.5-liteMSMARCO· Agentset22.3%
5Gemini Embedding 2MSMARCO· Agentset22.1%
6voyage-3-largeMSMARCO· Agentset22.1%
7voyage-3.5MSMARCO· Agentset22.1%
8voyage-4MSMARCO· Agentset22.1%
9BGE M3MSMARCO· Agentset21.9%
10Jina Embeddings v3MSMARCO· Agentset21.9%
11Jina Embeddings v5 Text SmallMSMARCO· Agentset21.9%
12Embed Multilingual V3.0MSMARCO· Agentset21.8%
13Qwen3 Embedding 0.6BMSMARCO· Agentset21.5%
14Text Embedding 3 SmallMSMARCO· Agentset21.2%
15Text Embedding 004MSMARCO· Agentset20.8%