EnterpriseRAG-Bench: 500,000+ Enterprise Documents and 500 Questions for RAG Evaluation

500,000+ documents and 500 questions form a benchmark for evaluating Retrieval-Augmented Generation systems on realistic enterprise data. The dataset was created by onyx-dot-app and last updated on 2026-04-21. It is described as the first publicly accessible dataset focused on company-internal documents, contrasting with existing benchmarks that use public web data.

Use Cases

• Benchmarking RAG system performance based on the 500 provided questions and enterprise document corpus
• Developing retrieval models for internal company knowledge bases based on the realistic enterprise data focus
• Evaluating the robustness of question-answering pipelines on domain-specific, non-public data mentioned in the description

Strengths

• Contains over 500,000 documents, providing a substantial corpus for testing
• Includes 500 specific questions for systematic evaluation
• Focuses on realistic enterprise data, a domain described as underserved by existing public benchmarks

Limitations

• Column-level documentation is absent; field semantics must be inferred after download
• Row count for the document corpus is unknown, which may limit suitability assessment
• Description metadata is limited; actual data quality requires manual inspection after download

Provenance

Source

onyx-dot-app on Hugging Face

Collection Method

Methodology and tools are referenced in a linked GitHub repository; the description suggests it was compiled to address a gap in enterprise-focused benchmarks.

Freshness

Last updated 2026-04-21 19:34:55; freshness should be verified