Jina AI, part of Elastic, has released jina-ocr-v1, an end-to-end visual document parser. It accepts PDFs, scans, tables, charts, or invoices and returns clean Markdown in a single pass. The model has 3.4B total parameters, with approximately 570M decoder parameters active per token. A speculative decoding head ships inside the checkpoint. Jina AI designed it to serve on low-budget GPUs such as the NVIDIA L4.
The technical report lists 91.14 on OmniDocBench v1.6 and 83.4 on olmOCR-Bench, placing it among the stronger open document parsers heading into 2026. As agentic AI pipelines increasingly need reliable, low-cost document ingestion, models like this are becoming infrastructure rather than research curiosities.
Is it deployable? Yes, for research and non-commercial use. The open weights are about 6.8 GB in BF16 and run on Transformers or vLLM. The CC BY-NC 4.0 license means commercial use requires contacting Jina AI.
What is jina-ocr-v1?
The model post-trains DeepSeek-OCR and retains its two efficiency components. DeepEncoder has about 380M parameters and chains SAM, a 16x convolutional compressor, and CLIP-L. It converts a 1024Γ1024 page view from 4,096 patches 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. Top-6 routing activates about 570M parameters per token. The position limit is 32,768. Output is Markdown, with tables rendered as HTML and formulas as LaTeX.
How FastMTP Speculative Decoding Works
OCR output is near-deterministic and locally structured, which makes it a strong fit for speculative decoding. Jina AI adds a FastMTP head: a single dense draft block applied recursively for K=3 steps. Draft parameters stay constant as depth grows, keeping the overhead predictable regardless of how many tokens are speculated.
This is particularly relevant in 2026, as document-parsing workloads scale across agentic RAG systems, compliance pipelines, and multimodal data extraction tools where GPU budgets remain tight. By embedding speculative decoding directly into the checkpoint rather than requiring a separate draft model, jina-ocr-v1 reduces deployment complexity and memory footprintβboth key constraints for teams running inference on L4-class hardware.
The release signals a broader trend: document AI is moving toward compact, efficient, single-pass architectures that can operate at the edge of enterprise infrastructure without sacrificing output fidelity.
via MarkTechPost