BenchmarkSciFact

BEIR SciFact

A scientific fact retrieval dataset in the BEIR benchmark, containing 5,183 documents and 300 queries. It is used to evaluate the retrieval quality of text embedding models.

Timeline (last 90 days)

Oct 13

In testing on the full BEIR SciFact dataset (5,183 documents, 300 queries), Qwen3-Embedding-8B ranked first with 0.7953 nDCG@10, with EmbeddingGemma-300M close behind at 0.7861.

Unverified50%

All Facts (1)

Unverified

In testing on the full BEIR SciFact dataset (5,183 documents, 300 queries), Qwen3-Embedding-8B ranked first with 0.7953 nDCG@10, with EmbeddingGemma-300M close behind at 0.7861.

50%

Source Articles

Local Embedding Model Selection Guide: Hardware Requirements and Cost AnalysisOct 13