Jina-OCR-v1: Efficient Document Parsing with Speculative Decoding and Dense Verifiable Rewards

arXiv:2609.03181 · cs.CL, cs.CV · Submitted 2026-09-02 · Read on arXiv

cs.CL, cs.CV

Submitted: 2026-09-02

Updated: 2026-09-02

Comments: 15 pages, 5 figures, 8 tables. Model at https://huggingface.co/jinaai/jina-ocr-v1

Code: https://github.com/rednote-hilab/dots.ocr

License: http://creativecommons.org/licenses/by-nc-sa/4.0/

The gist: We present Jina-OCR-v1, an end-to-end document parsing model built to serve on low-budget GPUs.

Terminology

Abstract

We present Jina-OCR-v1, an end-to-end document parsing model built to serve on low-budget GPUs. It combines the compressed-vision encoder and the 3B mixture-of-experts decoder of DeepSeek-OCR, which activates about 570M parameters per token, with a FastMTP speculative decoding head that shares a single draft block recursively across K=3 prediction steps. Greedy verification makes decoding lossless. Post-training combines instruction alignment, robustness fine-tuning on difficult documents, and GRPO under dense verifiable rewards: deterministic formula, table, and structural checks that award partial credit. The training data mixes cleaned public corpora with targeted synthetic pages. At the default dynamic-resolution setting, Jina-OCR-v1 scores 91.14 on OmniDocBench v1.6 and 83.4 on olmOCR-Bench, and reaches the highest page throughput in our comparison at 2.57 pages per second. On a low-budget GPU such as the NVIDIA L4, FastMTP doubles decoding speed over greedy autoregressive decoding. The model is publicly available at https://huggingface.co/jinaai/jina-ocr-v1.

Sources

Related papers