CLAP NQ: Benchmark Long-form Question Answering Dataset for RAG

CLAP NQ is a benchmark dataset for evaluating the full Retrieval-Augmented Generation pipeline. It provides long answers grounded in gold passages from the Natural Questions dataset, with answers described as concise and cohesive. The dataset was created by PrimeQA and last updated on May 13, 2026.

Use Cases

• Benchmarking retrieval models based on the provided corpus of gold passages
• Evaluating text generation models based on the annotated long-form answers
• Testing full RAG pipeline performance using the integrated retrieval and generation components
• Studying answer conciseness and cohesion based on the described answer properties

Strengths

• Answers are described as concise, being 3x smaller than the full source passage
• Answers are described as cohesive, integrating multiple non-contiguous pieces of the source passage
• Provides a full corpus for evaluating retrieval, generation, or the complete RAG pipeline

Limitations

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

Provenance

Source

Natural Questions (NQ) dataset

Collection Method

Derived and annotated from the source dataset

Freshness

Last updated 2026-05-13 12:56:33; freshness should be verified