Case Study: European Patent Office achieves faster, near-error-free patent transcription with Mistral AI

European Patent Office transcribes 400,000 pages a day with Mistral AI

The European Patent Office (EPO) faced the challenge of accurately processing millions of highly technical patent pages annually. Standard OCR tools struggled with complex elements like formulas and tables, failing to meet the required ST36 XML compliance and near-zero fault tolerance. To modernize its operations, the EPO partnered with vendor Mistral AI to implement a specialized Optical Character Recognition (OCR) solution.

Mistral AI deployed a tailored OCR model, fine-tuned on over 150,000 patent documents, which converted complex PDFs into structured ST36 XML format. The solution achieved a throughput of 400,000 pages per day with a character error rate of less than 1%, reducing processing lead time from days to minutes. This partnership with Mistral AI provided the EPO with a scalable, fault-tolerant system that improved efficiency and allowed experts to focus on high-value examination tasks.

European Patent Office