# Jina AI releases jina-ocr-v1 3.4B MoE document parser for low‑budget GPUs

Digest AI · Generative AI & Models · published 2026-09-18T21:21:59Z

Canonical: https://digestai.news/story/jina-ai-releases-jina-ocr-v1-3-4b-moe-document-parser-for-lowbudget-gp

## Summary

Jina AI, part of Elastic, announced the open‑weight release of jina-ocr-v1, a visual document parser that converts PDFs, scans, tables, charts or invoices into clean Markdown in a single pass. The model contains 3.4 B total parameters, with about 570 M decoder parameters active per token thanks to a 64‑expert MoE architecture and a speculative decoding head built into the checkpoint. It runs on low‑budget GPUs such as the NVIDIA L4 and is packaged in a 6.8 GB BF16 weight file.

On benchmark tests the model scores 91.14 on OmniDocBench v1.6 and 83.4 on olmOCR‑Bench, trailing only higher‑scoring systems like PaddleOCR‑VL‑1.6. Throughput reaches 2.57 pages per second on a single A100 40 GB GPU, the fastest among 14 systems measured, and up to 185.6 tokens per second on an NVIDIA L4 with CUDA graphs. The model is released under a CC BY‑NC 4.0 license, allowing research and non‑commercial use, while commercial users must contact Jina AI. It can be accessed via the Jina Reader header, an OpenAI‑compatible endpoint, or by self‑hosting with Transformers or vLLM.

## Key points

- jina-ocr-v1 has 3.4 B parameters, about 570 M active per token via a 64‑expert MoE decoder.
- Scores 91.14 on OmniDocBench v1.6 and 83.4 on olmOCR‑Bench; parses 2.57 pages / s on a single A100 GPU.
- Released under CC BY‑NC 4.0; usable via Jina Reader header, OpenAI‑compatible API, or self‑hosting with weights ~6.8 GB BF16.

## Why it matters

The model delivers high‑accuracy OCR with fast throughput on inexpensive GPUs, lowering the cost barrier for businesses and developers needing document parsing.

## Sources

1. [Jina AI Releases jina-ocr-v1: A 3.4B MoE Document Parser With Built-In Speculative Decoding for Low-Budget GPUs](https://marktechpost.com/2026/09/18/jina-ai-releases-jina-ocr-v1-a-3-4b-moe-document-parser-with-built-in-speculative-decoding-for-low-budget-gpus) (MarkTechPost, 2026-09-18)

## Cite

Digest AI, "Jina AI releases jina-ocr-v1 3.4B MoE document parser for low‑budget GPUs", 18 September 2026, https://digestai.news/story/jina-ai-releases-jina-ocr-v1-3-4b-moe-document-parser-for-lowbudget-gp

---

Written by Digest AI's editorial model from the linked sources; the sources are the record. Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse
JSON: https://digestai.news/story/jina-ai-releases-jina-ocr-v1-3-4b-moe-document-parser-for-lowbudget-gp.json