Xl Docbench

1,519 questions comprise this benchmark for evaluating long-context, multimodal, and cross-document understanding. The dataset, created by 'anonymous12123' and last updated in May 2026, includes benchmark questions, answers, public source URLs for documents, and human-annotated evidence pages and snippets. It contains 331 public document records in a separate file.

Use Cases

• Benchmarking model performance on long-context question answering based on the 1,519 provided questions.
• Evaluating multimodal document understanding capabilities using the human-annotated evidence pages and snippets.
• Testing cross-document reasoning and information synthesis across the 331 provided public documents.
• Training retrieval-augmented generation (RAG) systems using the provided document URLs and evidence annotations.

Strengths

• Includes 1,519 benchmark questions specifically for long-context and multimodal understanding.
• Provides human-annotated evidence pages and snippets for evaluation.
• Contains 331 public document records with source URLs and metadata.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the primary QA file is unknown, which may limit suitability assessment.
• The author and organization are listed as 'anonymous12123' and unknown, respectively, limiting provenance clarity.

Provenance

Source

huggingface

Collection Method

Likely manually curated and annotated for benchmarking purposes.

Time Range

null

Freshness

Last updated 2026-05-07 11:19:43; freshness should be verified.

Geography

null