Jina AI has released jina-ocr-v1, a document parsing model that converts PDFs, scanned files, tables, and charts directly into Markdown. The company said the model was post-trained on top of DeepSeek-OCR and keeps its mixture-of-experts design, with roughly 3.4 billion total parameters and about 570 million parameters activated per token during decoding. Jina also added FastMTP speculative decoding.
In Jina’s own tests, jina-ocr-v1 scored 91.14 on OmniDocBench v1.6, 0.89 points above DeepSeek-OCR-2. On olmOCR-Bench, it posted 83.4, which Jina said was 7.4 points higher than the DeepSeek-OCR base model it actually builds on. Throughput was a central part of the release. Under a single NVIDIA A100 with concurrency set at 32, Jina said the model reached 2.57 pages per second, the highest among 14 systems it tested, versus 2.10 pages per second for DeepSeek-OCR, or about 22% higher. The model weights are already available on Hugging Face under a CC BY-NC 4.0 license, and commercial use requires contacting Jina.
Jina AI has released jina-ocr-v1, a document parsing model that can turn PDFs, scanned documents, tables, and charts directly into Markdown.
Architecture and training setup
According to Jina AI, jina-ocr-v1 was post-trained on top of DeepSeek-OCR. It keeps the same mixture-of-experts architecture, with about 3.4 billion total parameters and roughly 570 million parameters activated per token during decoding, and adds FastMTP speculative decoding.
Benchmark scores
In Jina’s self-run tests, jina-ocr-v1 scored 91.14 on OmniDocBench v1.6, 0.89 points above DeepSeek-OCR-2. On olmOCR-Bench, it scored 83.4, which Jina said was 7.4 points higher than the DeepSeek-OCR base model it actually uses.
Throughput figures
Throughput was one of the main points highlighted by Jina. Under a single A100 GPU with concurrency at 32, the model reached 2.57 pages per second, the highest among the 14 systems tested by Jina. That compares with 2.10 pages per second for DeepSeek-OCR, or about 22% higher.
Availability and license
Jina said the model weights have been uploaded to Hugging Face under a CC BY-NC 4.0 license. Commercial use requires contacting Jina.
This article was originally published by Bit.Fan. For more cryptocurrency news and market insights, visit www.bit.fan. Disclaimer:
The market information, project data, and third-party content displayed on this platform are for industry information sharing only and do not constitute any form of investment advice or return commitment.
Cryptocurrency trading carries high risks. Users should fully assess their risk tolerance and make independent decisions. All profits, losses, and legal responsibilities are borne by the users themselves.
