OCRBench v2
A large-scale bilingual (English/Chinese) text-centric benchmark of ~10,000 human-verified QA pairs across 31 scenarios that evaluates large multimodal models on visual text localization, recognition, parsing, and reasoning.
What this benchmark measures
A large-scale bilingual (English/Chinese) text-centric benchmark of ~10,000 human-verified QA pairs across 31 scenarios that evaluates large multimodal models on visual text localization, recognition, parsing, and reasoning.
Rows on this page are sourced from public benchmark artifacts, leaderboard exports, or source-linked model reports. Each row keeps benchmark version, source model name, and available run details attached to the score.
The metric shown here is accuracy. It should be interpreted within OCRBench v2, not compared as part of a site-wide ranking.
Frequently asked
What is OCRBench v2?
A large-scale bilingual (English/Chinese) text-centric benchmark of ~10,000 human-verified QA pairs across 31 scenarios that evaluates large multimodal models on visual text localization, recognition, parsing, and reasoning. It is a multimodal benchmark measured by accuracy.
What does accuracy mean on OCRBench v2?
OCRBench v2 reports accuracy; higher is better. Scores are shown only within OCRBench v2 and are never averaged with other benchmarks.
What is the top reported OCRBench v2 score?
Gemini 3 Pro has the top reported score on OCRBench v2: 63.4 (accuracy).
Why do OCRBench v2 scores differ across runs?
Harness, scaffold, reasoning effort, and prompt setup change results, so two runs of the same model can differ. evals.report keeps each score with its run context so the differences stay visible.
Does evals.report rank models across benchmarks?
No. OCRBench v2 scores are shown within their own metric; evals.report never combines benchmarks into a composite ranking or a single "best model".