Jina AI Launches jina-ocr-v1: A 3.4B MoE Document Parser
Jina AI introduces jina-ocr-v1, a visual document parser with 3.4 billion parameters and built-in speculative decoding for low-budget GPUs.
What happened
Jina AI released jina-ocr-v1, an end-to-end visual document parser that converts PDFs, scans, tables, charts, and invoices into clean Markdown text in one pass. The model has 3.4 billion total parameters with about 570 million active per token.
Why it matters
jina-ocr-v1 is optimized for low-budget GPUs like the NVIDIA L4 and includes a FastMTP speculative decoding head that drafts three tokens per step while maintaining lossless output. It scores 91.14 on OmniDocBench v1.6 and 83.4 on olmOCR-Bench, making it highly accurate.
For builders
The model is available for research and non-commercial use under a CC BY-NC 4.0 license. Weights are about 6.8 GB in BF16 format and can run on either Transformers or vLLM frameworks.
Try this
Read the source before changing your stack to understand how jina-ocr-v1 integrates speculative decoding for efficient performance.
Back to builder briefs