Skip to content

Benchmark

Agentset Embedding Win Rate - SciFact

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
0.34

Score distribution

48.5%
Mean
44.4%
Median
12.1%
Std dev
32.5%
10th percentile
62.2%
90th percentile
29.4%
Min
29.4%70.6%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.34 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· Agentset70.6%
2voyage-4SciFact· Agentset62.2%
3Text Embedding 3 LargeSciFact· Agentset61.3%
4Jina Embeddings v5 Text SmallSciFact· Agentset59.4%
5Embed Multilingual V3.0SciFact· Agentset57.5%
6Qwen3 Embedding 0.6BSciFact· Agentset55.6%
7voyage-3-largeSciFact· Agentset55%
8Text Embedding 3 SmallSciFact· Agentset44.4%
9Text Embedding 004SciFact· Agentset43.8%
10Jina Embeddings v3SciFact· Agentset43.1%
11Qwen3 Embedding 8BSciFact· Agentset41.3%
12voyage-3.5-liteSciFact· Agentset36.3%
13voyage-3.5SciFact· Agentset35.6%
14Qwen3 Embedding 4BSciFact· Agentset32.5%
15BGE M3SciFact· Agentset29.4%