88Z AI, 29 October — ModelBest open-sources the MiniCPM-OCR series, with the 1.3B model achieving SOTA on two benchmarks and the 2B model being the industry's first OCR result discrimination model.
At a glance
- The 1.3B model achieved SOTA on two benchmarks with a throughput of 1.217 pages/sec after acceleration
- The 2B model is the industry's first OCR result discrimination model with an AUROC of 93.6
The update
ModelBest has officially released and open-sourced the MiniCPM-OCR series. The 1.3-billion parameter MiniCPM-OCR 1.0 achieved state-of-the-art results on both the Real5-OmniDocBench and OCRBench v2 benchmarks. After acceleration with DSpark, the model reaches a throughput of 1.217 pages per second. Additionally, the 2-billion parameter MiniCPM-OCR Jev is the industry's first model dedicated to discriminating OCR parsing results. With an AUROC of 93.6, it can automatically filter for credible results and select the best output from multiple models, thereby reducing the need for manual verification and improving overall efficiency and accuracy.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- 面壁智能Source published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.