Tasks and inputs
MiniCPM-OCR Jev is listed for Text classification, Document OCR. It accepts image, text and produces text. Use the publisher’s input format and preprocessing for your chosen deployment.
MiniCPM-OCR Jev is OpenBMB's open 2B judge that scores whether an OCR result for a text, formula or table region is correct, from 0 to 1.
The specification score summarizes recorded features and limits. It is not a benchmark result.
Pricing conditions: No API pricing stated; open weights for self-hosting
Recorded modality support shows what this model can accept and produce. An undocumented modality is shown as unknown rather than assumed to be unsupported.
Accepts
Produces
Every OCR pipeline eventually needs the same answer: which of these results can we trust without a human looking? MiniCPM-OCR Jev is a model for exactly that question. Give it a region of a document and a candidate transcription, and it returns a score between 0 and 1. Higher means more likely correct.
OpenBMB open-sourced it on 27 October 2026, alongside MiniCPM-OCR 1.0, under Apache 2.0.
Jev is trained from Qwen3.5-2B. The vision encoder and language backbone stay; the output head is replaced with a binary classifier. It looks at three things together:
Comparing the rendered candidate with the real crop is what makes it good at catching subtle errors in formulas and tables. If you call the hosted API, the Docker image or the SDK with its built-in renderer, you only send the crop, the text and the region type (text, formula or table); rendering is handled for you.
On JevOCRBench, OpenBMB reports an AUROC of 93.6 (×100). For comparison in the same table, a 35B-A3B judge scores 90.1 and Qwen3.5-397B-A17B scores 93.8. Across outputs from four different OCR models (MiniCPM-OCR 1.0, GLM-OCR, PaddleOCR-VL-1.6 and Xiaomi-OCR-0) it stays between 91.9 and 93.4. A judge that works on other parsers' output is the useful kind.
| Use | How |
|---|---|
| Auto-accept | Pass regions above a threshold straight through |
| Human review | Route low scores to an annotator queue |
| Ensemble | Run several OCR engines and keep the highest-scoring result |
Do not take a default. Label a few hundred regions from your own documents as correct or not, score them, and choose the threshold that matches your tolerance for errors slipping through versus review cost. Recheck it whenever you change OCR engines. Pair it with any parser from the document OCR directory; smaller text models from the same lab, like MiniCPM5 2B, are not substitutes, because Jev's head is a classifier, not a generator.
Compare the model on the work you need to complete.
MiniCPM-OCR Jev is listed for Text classification, Document OCR. It accepts image, text and produces text. Use the publisher’s input format and preprocessing for your chosen deployment.
Measure a completed task using the prompts, documents and tools your application will use. Include reasoning and retries in the result.
The weights are downloadable under Apache-2.0. Check runtime support and size the complete model and working context for your expected concurrency.
These records share a directory category. Check the source card, output type and deployment requirements before treating a model as a replacement.
OpenBMB
MiniCPM-OCR 1.0 is OpenBMB's open 1.3B document-parsing model for text, tables, formulas and charts, with a free API and on-device apps.
Tencent
Youtu-Parsing-Omni is Tencent Youtu's 5B open parser that turns documents, charts, images, audio and video into one structured JSON format.
lightonai
LightOnOCR-3-0.8B is LightOn's smallest open document model: page images in, markdown out, with optional bounding boxes and chart tables.
Sarvam
Sarvam Vision 2.1 is Sarvam's 3B document-intelligence VLM for OCR, tables, forms and Indic handwriting in 22 Indian languages plus English.
NVIDIA
Nemotron 3.5 Parse is NVIDIA's 0.631B open document parser that turns page images into Markdown, HTML or JSON with tables, charts and reading order.
Cohere
Cohere Parse (parse-v5.0) is a 2.3B vision-language document parser that turns PDFs, slides and scans into Markdown for RAG, at $1.50 per 1,000 pages.