ParseBench: A Benchmark for Evaluating Document Parsing Systems

ParseBench is a benchmark for evaluating document parsing systems on real-world enterprise documents. It was created by llamaindex and last updated on April 10, 2026. The benchmark is stratified into five capability dimensions: tables, charts, content faithfulness, semantic formatting, and visual grounding.

Use Cases

• Benchmarking table extraction algorithms based on the 'tables' evaluation dimension
• Evaluating chart understanding models based on the 'charts' evaluation dimension
• Testing content faithfulness of parsing outputs based on the 'content faithfulness' dimension
• Assessing semantic formatting recognition based on the 'semantic formatting' dimension
• Measuring visual grounding capabilities based on the 'visual grounding' dimension

Strengths

• Provides a multi-dimensional evaluation framework across five specific capability areas
• Focuses on real-world enterprise documents, which likely increases practical relevance
• Includes task-specific metrics designed to capture what agentic workflows depend on

Limitations

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

Provenance

Source

huggingface

Collection Method

Likely curated for benchmarking purposes by llamaindex.

Time Range

null

Freshness

Last updated 2026-04-10 14:52:34; freshness should be verified

Geography

null