Elastic ships jina-ocr-v1 document parser

Elastic released jina-ocr-v1, a layout-aware document parser available through its Inference Service and on-prem, handling tables, math, handwriting and over 100 languages in one model. It scores 83.4 on olmOCR-bench and 91.14 on OmniDocBench, above GPT-5.2 at 86.59, using 570 million active parameters from a 3.4 billion total mixture-of-experts design.