Elastic launches jina-ocr-v1 OCR model with 100+ language support and top benchmark scores

On September 17 2026 Elastic’s Search Labs blog introduced jina‑ocr‑v1, a new optical‑character‑recognition model designed to handle complex document layouts, tables, mathematical notation and more than 100 languages. The model contains roughly 570 million active parameters and, as reported by Elastic, achieves an 83.4 score on the olmOCR‑bench benchmark – the highest reported for any OCR model under 600 M parameters – and surpasses GPT‑5.2 on the OmniDocBench benchmark. Elastic positions the model as a single‑copy solution that can replace multiple specialized OCR engines in AI pipelines. The model is available through Elastic’s integration with Jina AI, though the performance figures are vendor‑provided and have not been independently verified.