Elastic (NYSE: ESTC) has launched jina-ocr-v1, an optical character recognition model designed to process complex documents in a single pass, according to a company press release.

The model uses a mixture-of-experts architecture with 3.4 billion total parameters and 574 million active parameters at inference. Elastic says this allows it to run at the speed and cost of a sub-600 million parameter model while delivering higher accuracy.

jina-ocr-v1 scores 83.4 on olmOCR-bench, which Elastic claims is the highest published score among models with fewer than 600 million active parameters. The company states the model is roughly one-tenth the size of the benchmark leader on that test.

The model supports more than 100 languages and can process scanned pages, photographed documents, slides, and files from formats including PDF, PPTX, and XLSX. It outputs structured Markdown, converts tables to basic HTML, reads handwriting including English cursive, and translates printed mathematical formulas into LaTeX.

"Traditional OCR pipelines break down with complex layouts, tables, handwriting and other highly visual content," said Han Xiao, vice president of AI at Elastic. "Until now, companies either had to accept those limitations or pay a significant premium to use general-purpose LLMs for ingesting documents."

jina-ocr-v1 is available through the Elastic Inference Service, included with Elastic Cloud, with GPU acceleration. Developers can also access it via the Jina API on a pay-per-token basis. On-premises deployment is available through commercial licensing from Elastic, or through Hugging Face under a CC BY-NC 4.0 license for academic and noncommercial use.