Skip to content

Benchmark

Agentset Embedding nDCG@10 - SciFact

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

Models measured
15
Scores recorded
15
Sources
1
IQ weight
1.34

Score distribution

71.7%
Mean
72.6%
Median
6.1%
Std dev
63.6%
10th percentile
77.9%
90th percentile
59.9%
Min
59.9%87.1%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.34 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
1Gemini Embedding 2SciFact· Agentset87.1%
2voyage-3-largeSciFact· Agentset77.9%
3voyage-4SciFact· Agentset75.8%
4Qwen3 Embedding 8BSciFact· Agentset74.4%
5Jina Embeddings v5 Text SmallSciFact· Agentset73.4%
6Text Embedding 004SciFact· Agentset72.9%
7Text Embedding 3 LargeSciFact· Agentset72.7%
8voyage-3.5-liteSciFact· Agentset72.6%
9voyage-3.5SciFact· Agentset70.5%
10Embed Multilingual V3.0SciFact· Agentset70.2%
11Qwen3 Embedding 4BSciFact· Agentset69.7%
12Text Embedding 3 SmallSciFact· Agentset68.4%
13Qwen3 Embedding 0.6BSciFact· Agentset66.6%
14Jina Embeddings v3SciFact· Agentset63.6%
15BGE M3SciFact· Agentset59.9%