jina-ocr-v1

jina-ocr-v1 is an efficient, end-to-end document parsing model designed to convert images and PDFs into structured Markdown in a single pass. It utilizes a 3.4B parameter mixture-of-experts architecture with 570M active parameters to deliver high-performance document transcription on low-budget hardware.

jina-ocr-v1 is an efficient, end-to-end document parsing model designed to convert images and PDFs into structured Markdown in a single pass. It utilizes a 3.4B parameter mixture-of-experts architecture with 570M active parameters to deliver high-performance document transcription on low-budget hardware.

• 01Lossless speculative decoding via FastMTP for accelerated inference
• 02Efficient 3.4B parameter MoE architecture with 570M active parameters
• 03Native Markdown output for tables, formulas, and text
• 04High-throughput performance reaching 2.57 pages per second
• 05Dynamic resolution support for varied document layouts