jina-ocr-v1

Name: jina-ocr-v1
Author: Jina AI

Jina AI · released 2026-09-02 · data checked 2026-11-06

jina-ocr-v1 is a 3.4B-parameter OCR and document-parsing model from Jina AI (DE), released 2026-09-02 under the CC-BY-NC-4.0 license. It scores 91.14 on OmniDocBench v1.6 (#24 of 33) and 83.4 on olmOCR-Bench (#8 of 24). It needs about 7.7 GB of memory at 16-bit.

Builds on DeepSeek-OCR (3B Mixture-of-Experts decoder, about 570M active) with faster speculative decoding.

OmniDocBench v1.6

91.14

#24 of 33 models

olmOCR-Bench

83.4

#8 of 24 models

Parameters

3.4B

570M active · Mixture-of-Experts

Memory to run

~7.7 GB

16-bit weights plus about 1 GB

Benchmarks

OmniDocBench v1.6 Lab-reported91.14

Best: OXR 1.0 97.03

olmOCR-Bench Lab-reported83.4

Best: OXR 1.0 87.8

The tick marks the best score on each benchmark. See how to read the ranking.

Specifications

Lab: Jina AI (DE)
Type: OCR / document parsing specialist
Total parameters: 3.4B
Active parameters: 570M
License: CC-BY-NC-4.0
Non-commercial only; commercial use needs a license from Jina AI.
Memory to run: ~7.7 GB at 16-bit
Released: 2026-09-02

Weights and usage on Hugging Face ↗

Closest alternatives

dots.ocr (3B, 90.77 on OmniDocBench, MIT)
MonkeyOCR v2 (B) (880M, 91.80 on OmniDocBench, Apache-2.0)
Infinity-Parser2 Flash (2.2B, 91.98 on OmniDocBench, Apache-2.0)
DeepSeek-OCR 2 (3.4B, 90.25 on OmniDocBench, Apache-2.0)

Sources

https://huggingface.co/jinaai/jina-ocr-v1

Something wrong here? Report it.