> ## Content Index
> Fetch the complete content index at: https://globalfeed.ai/llms.txt
> Use this file to discover other available public pages before exploring further.

# Jina AI's new document parser leads on throughput, not accuracy
- URL: https://globalfeed.ai/en/jina-ais-new-document-parser-leads-on-throughput-not-accuracy/
- Published: 2026-09-21T22:07:09.000Z
- Updated: 2026-09-21T22:07:09.000Z
- Description: Jina AI has released jina-ocr-v1, a document parser that converts PDFs and scans to Markdown in one pass and tops throughput rather than accuracy charts.
- Author: GlobalFeed Editor
- Tags: Jina AI, OCR, open weights, x-JinaAI_, dil-en

Jina AI, part of Elastic, has released **jina-ocr-v1**, an end-to-end visual document parser. It takes PDFs, scans, tables, charts or invoices and returns Markdown in a single pass, with tables in HTML and formulas in LaTeX. The model holds 3.4B total parameters, of which roughly 570M decoder parameters are active per token.

The open weights are about 6.8 GB in BF16 and run on Transformers or vLLM. The CC BY-NC 4.0 license limits use to research and non-commercial work; commercial deployment requires contacting Jina AI.

## Architecture

The model post-trains DeepSeek-OCR. Its DeepEncoder, around 380M parameters, chains SAM, a 16x convolutional compressor and CLIP-L, turning 4,096 patches from a 1024×1024 page view into 256 visual tokens. A dynamic-resolution mode adds up to 9 local tiles at 100 tokens each, capping a page at 1,156 visual tokens. The decoder is DeepSeek-3B-MoE with 12 layers, 64 routed experts and 2 shared experts, and a position limit of 32,768.

A speculative decoding head ships inside the checkpoint. The FastMTP head applies one dense draft block recursively for K=3 steps, and the decoder verifies the drafts greedily. Committed text always matches plain greedy decoding, so the speedup is lossless. At K=3 the model commits 2.73 tokens per step on average.

## Scores and speed

The technical report lists 91.14 on OmniDocBench v1.6 and 83.4 on olmOCR-Bench. The model does not lead on accuracy: PaddleOCR-VL-1.6 and HunyuanOCR-1.5 (94.74) score higher on OmniDocBench, while chandra-ocr-2 and dots.mocr (83.9) score higher on olmOCR-Bench. Post-training adds 7.4 points over the DeepSeek-OCR backbone on olmOCR-Bench.

Throughput is where it stands out. On one A100 40 GB at concurrency 32, jina-ocr-v1 parses 2.57 pages per second, the highest among the 14 systems Jina AI measured, against 1.22 for olmOCR-2 and 0.38 for chandra-ocr-2\. It emits 1,085 output tokens per page. On an NVIDIA L4 at batch size 1, eager decoding rises from 42.7 to 83.1 tokens per second, a 1.95x speedup at a 57.6% acceptance rate.

The fastest way to test it is Jina Reader: send a URL to r.jina.ai with the header X-Respond-With: jina-ocr-v1\. Jina AI also hosts an OpenAI-compatible endpoint and a demo. Full details are in the [MarkTechPost report](https://www.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/?ref=globalfeed.ai).