Understand this story
The English summary is being updated. See the reporting timeline below.
Latest development · Report
2026 OCR Model Rankings for Tables and Charts: OCRia-1 Leads in Structured ParsingDevelopments
1 confirmed developments
- Source:English text is being prepared.
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Nov 18
LlamaIndex Products, Engineering, and EvaluationsFirst-partySelected
2026 OCR Model Rankings for Tables and Charts: OCRia-1 Leads in Structured Parsing
LlamaIndex ranks several OCR models and APIs, concluding that OCRia-1, a 4B model running on LlamaParse, is best suited for tables, charts, and complex structured documents. OCRia-1 scores 91.5 on the table dimension and 85.1 on the chart dimension of ParseBench. For plain-text scenarios, AWS Textract, Google Document AI, Azure Document Intelligence, or open-weight models can be selected based on price, language coverage, and deployment method. The article also discloses that ParseBench contains approximately 2,000 manually verified pages and more than 169,000 test rules, and cross-checks the table results against three external benchmarks.
AI Trends