OCRBench: Results as Published

OCRBench tests whether a multimodal model can read text in an image and answer from it. The scored set has 1,000 manually checked question-answer pairs in five parts: text recognition, scene-text visual question answering, document-oriented visual question answering, key information extraction, and handwritten mathematical expression recognition. Those pairs come out of a wider study that uses 29 text datasets. The published result is a final score that adds the five part scores.

Every number here is read from the same catalog file as the benchmark table, data version 2026103001, and is the number the table shows for that model and test. Where one page printed a model under several setups, each value is listed. The catalog is public in armanobosyan/sugra-llm-benchmarks; a wrong number can be reported from the table, with Report an error under the value, or as an issue in that repository.