Datasets

Available benchmarks for RAG evaluation.

Comparison

Dataset	Modality	Queries	Documents	Generation GT	Use
BEIR	Text	300-15k	3k-5M	No	Retrieval
MTEB	Text	varies	varies	No	Retrieval
RAGBench	Text	varies	varies	Yes	RAG
MrTyDi	Text	varies	varies	No	Multilingual
BRIGHT	Text	varies	varies	No	Retrieval
Open-RAGBench	Multimodal	varies	varies	Yes	RAG
ViDoRe	Multimodal	varies	varies	No	Visual
ViDoRe v2	Multimodal	varies	varies	No	Visual
ViDoRe v3	Multimodal	varies	varies	No	Visual
VisRAG	Multimodal	varies	varies	No	Visual

Text vs Multimodal

Text datasets:

• Documents as plain text
• Text embeddings + BM25 tokens
• Use: Text Retrieval Tutorial

Multimodal datasets:

• Documents as images (PDF pages)
• Image embeddings (ColPali)
• Use: Multimodal Tutorial

Browse

• Text Datasets
• Multimodal Datasets