MiniCPM-OCR pairs recognition with a checker model
The pairing of a parser with a dedicated verifier targets the common failure where OCR succeeds but returns wrong text, giving enterprises a risk-based review workflow rather than blind trust in extraction. Since Jev is model-agnostic, teams already running PaddleOCR or GLM-OCR can add quality scoring without switching pipelines. The claimed 128 pages per minute throughput makes it practical for batch document processing workloads.
OpenBMB released MiniCPM-OCR, a two-model OCR system that combines document parsing with quality checking. MiniCPM-OCR 1.0 is a 1.3B model that recognises text, tables and formulas from scanned PDFs or document photos and outputs structured content. MiniCPM-OCR Jev, a 2B model, scores the reliability of OCR results and flags errors. With DSpark acceleration, the 1.0 model reaches 1.49x the throughput of PaddleOCR-VL-1.6 and 3.62x that of MinerU2.5 Pro, and with SDK concurrency optimisation the vendor claims up to 128 pages per minute. Jev works independently of MiniCPM and can also validate output from GLM-OCR and PaddleOCR.
OpenBMB刚上了一套OCR模型：MiniCPM-OCR，它把 OCR识别和OCR质量检查结合了起来，不仅会解析文档，还得判断自己有没有识别错
解决OCR成功但不一定识别正确的问题
两个模型组合，一个干活，一个质检 MiniCPM-OCR 1.0，1.3B，负责识别和解析文档，小、高精度、高速度 MiniCPM-OCR Jev，2B，专门检查OCR结果可靠性
输入扫描件PDF或文档照片，MiniCPM-OCR 1.0识别文字、表格、公式，输出结构化内容，Jev给结果质量打分
MiniCPM-OCR 1.0用DSpark加速后，吞吐量能到 PaddleOCR-VL-1.6的1.49倍、MinerU2.5 Pro的3.62倍。结合SDK并发优化，说是最高每分钟能并发处理128页
MiniCPM-OCR Jev不一定非得配MiniCPM，也能检查 GLM-OCR、PaddleOCR等的结果
OCR解析+AI质检+按风险复核这套工作流，应该是更接近企业使用场景的
#OCR #MiniCPM #AI文档解析
