Jina ships 3.4B OCR model for low-budget GPUs

Jina AI released jina-ocr-v1, a 3.4-billion-parameter document parser with roughly 570 million active decoder parameters that reports 2.57 pages per second on one A100. Jina’s release notes report 91.14 on OmniDocBench and 83.4 on olmOCR-Bench, while its model card says the weights are CC BY-NC 4.0, so commercial use requires contacting Jina. Teams replacing hosted PDF, table, or invoice extraction can benchmark it locally, but should test field and table accuracy—not only throughput—and price the license before treating an open checkpoint as a commercial drop-in; MLPerf’s RAG analysis makes the same throughput-versus-acceptance distinction.