Document Understanding Benchmark

OCR Bench v2

OCRBench v2 is the second-generation OCR/document-understanding benchmark suite with harder reading and parsing tasks. Recorded from the Bonsai 2 27B GGUF card's thinking-mode evaluation (EvalScope + vLLM on NVIDIA H100); treated as distinct from plain OCRBench.

Category

Document Understanding

Score type

percent

Scale

100%

higher is better

Models tested

2

What is OCR Bench v2?

OCR Bench v2 is a benchmark for reading and parsing real documents. The official description says: "OCRBench v2 is the second-generation OCR/document-understanding benchmark suite with harder reading and parsing tasks". Below we describe in simple words what benchmarks of this type generally examine.

What does it test?

OCR Bench v2 measures reading and parsing real documents. In detail, it checks:

  • extracting structured text from scans, PDFs and photos
  • keeping tables, headings and reading order intact
  • answering comprehension questions about the document content
  • handling noisy layouts, handwriting and multi-page documents

How to read the score

The score is accuracy: the percentage of tasks answered or solved correctly. 100% means every task was solved perfectly; higher is better.

Official methodology note: OCRBench v2 is the second-generation OCR/document-understanding benchmark suite with harder reading and parsing tasks. Recorded from the Bonsai 2 27B GGUF card's thinking-mode evaluation (EvalScope + vLLM on NVIDIA H100); treated as distinct from plain OCRBench.

Frequently asked questions

What is the OCR Bench v2 benchmark?

OCR Bench v2 is a benchmark for reading and parsing real documents. The official description says: 'OCRBench v2 is the second-generation OCR/document-understanding benchmark suite with harder reading and parsing tasks'. Below we describe in simple words what benchmarks of this type generally examine.

What does OCR Bench v2 test?

OCR Bench v2 mainly measures reading and parsing real documents. Concretely, it checks: extracting structured text from scans, PDFs and photos; keeping tables, headings and reading order intact; answering comprehension questions about the document content.

How is the OCR Bench v2 score calculated?

The score is accuracy: the percentage of tasks answered or solved correctly. 100% means every task was solved perfectly; higher is better. The declared maximum score of this benchmark is 100.0.

Where do I get OCR Bench v2 data and scores?

The official benchmark page/reference is linked on this page. Every score on Open LLM Radar is also available as open data: download the scores JSON (http://open-llm-radar-api:8000/api/v1/benchmarks/ocr-bench-v2/scores) or pull it directly from the Open LLM Radar API with any HTTP client.

Model scores 2 models tested

Reset
# Model Score Date Source Verified
1 LFM3-VL-2B
LiquidAI
63.5%
— Report Verified
2 Bonsai 2 27B — GGUF
Prism ML
56.88%
17.09.2026 Report Verified

What now?

You've seen how models score on OCR Bench v2 — here's what to do next.