image to text benchmark
OCRBench_V2
OCRBench v2: Enhanced large-scale bilingual benchmark for evaluating Large Multimodal Models on visual text localization and reasoning with 10,000 human-verified question-answering pairs across 8 core OCR capabilities
Updated Aug 7, 2026
Published models7
Registry coverage7
MetricScore
EvidenceB
OCRBench_V2 leaderboard
Sorted by the source-provided rank. Higher score is better according to the registry.
| 1 | AC | 67.1% | 100.0% | 7 | C | |
| 2 | AM Nova 2 ProAmazon | 64.5% | 83.3% | 7 | C | |
| 3 | BY | 63.2% | 66.7% | 7 | C | |
| 4 | BY | 62.8% | 50.0% | 7 | C | |
| 5 | AMNova 2 OmniAmazon | 58.2% | 33.3% | 7 | C | |
| 6 | AC | 57.8% | 16.7% | 7 | C | |
| 7 | AMNova 2 LiteAmazon | 56.1% | 0.0% | 7 | C |
Score distribution
Top published rows on the benchmark's original scale.
OCRBench_V2 highlights
The top published results on this benchmark's own scale.
What is OCRBench_V2?
Definition and scoring fields from the benchmark registry.
OCRBench v2: Enhanced large-scale bilingual benchmark for evaluating Large Multimodal Models on visual text localization and reasoning with 10,000 human-verified question-answering pairs across 8 core OCR capabilities
Scores are shown in ratio. The current registry marks this benchmark as not independently verified with evidence level B.
- Family
- OCRBench_V2
- Modality
- multimodal
- Primary category
- image to text
- Score direction
- higher
- LLMBoard eligible
- No
- Evaluation key
- ocrbench-v2|llm-stats-current
Source-native results are preserved. Eligibility for the overall LLMBoard score is a separate policy decision.
FAQ
Common questions about OCRBench_V2.
Which model scores highest on OCRBench_V2?
Qwen3.7-Plus is currently ranked first with 67.1%.
What does OCRBench_V2 measure?
OCRBench v2: Enhanced large-scale bilingual benchmark for evaluating Large Multimodal Models on visual text localization and reasoning with 10,000 human-verified question-answering pairs across 8 core OCR capabilities
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
7 unique published model results are currently shown.
Does this benchmark affect the overall score?
This benchmark is preserved as source-native evidence but is not eligible for the current overall score.