# MiniCPM-OCR Jev 2B model scores OCR results with 93.68% accuracy on real-world benchmarks

Digest AI · Research · published 2026-10-27T16:21:51Z

Canonical: https://digestai.news/story/minicpm-ocr-jev-2b-model-scores-ocr-results-with-93-68-accuracy-on-rea

## Summary

**MiniCPM-OCR Jev (2B)**, a 2-billion-parameter model, achieves **93.68% AUROC** on real-world document parsing, outperforming larger models like **JEV-35B-A3B (90.1%)** and nearly matching **Qwen3.5-397B-A17B (93.8%)** on the **JevOCRBench** dataset. Developed by the same team behind **MiniCPM-OCR 1.0**, it evaluates OCR outputs for text, formulas, and tables by scoring candidate accuracy on a **0–1 scale**, enabling automated acceptance or human review of results based on a configurable threshold. The model retains its vision encoder and language backbone but replaces the output head with a binary classifier trained on **Qwen3.5-2B** data, ensuring stable generalization across different OCR parsers (e.g., **GLM-OCR, PaddleOCR-VL-1.6, Xiaomi-OCR-0**), with scores ranging from **91.9–93.4%** on their outputs. Deployment requires users to supply region images, candidate text, and category labels (text/formula/table), with optional built-in rendering or manual pre-processing. The model’s **AUROC metric** measures ranking performance, not absolute accuracy, and users must empirically determine thresholds for their specific error tolerances and document types. Downloads for the model saw **2 downloads last month** on its hosting platform, per the post’s metrics.

## Key points

- MiniCPM-OCR Jev 2B scores OCR results with **93.68% AUROC** on JevOCRBench, surpassing larger models like JEV-35B-A3B (90.1%)
- Model evaluates text, formulas, and tables by comparing candidate text to region images, enabling automated filtering or human review
- Deployment requires threshold tuning based on error budgets and document types; built-in rendering or manual pre-processing is optional

## Why it matters

This model reduces manual OCR review workloads by automating quality checks for text, formulas, and tables, improving accuracy in document processing pipelines. Its small size (2B params) and high performance make it a cost-effective alternative to larger models for enterprises handling high-volume OCR tasks.

## Sources

1. [A 1.3B OCR model just hit 93.68 on real-world document parsing and there's now an AI judge for OCR errors](https://huggingface.co/openbmb/MiniCPM-OCR-Jev) (huggingface.co, 2026-10-27, primary source)

## Cite

Digest AI, "MiniCPM-OCR Jev 2B model scores OCR results with 93.68% accuracy on real-world benchmarks", 27 October 2026, https://digestai.news/story/minicpm-ocr-jev-2b-model-scores-ocr-results-with-93-68-accuracy-on-rea

---

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/minicpm-ocr-jev-2b-model-scores-ocr-results-with-93-68-accuracy-on-rea.json