AI inside Releases “PolySphere-3,” a Major Update of Its In-House Developed LLM
Tokyo, June 4, 2025 – AI inside Inc., a provider of an AI platform, announces the major update of its in-house developed Large Language Model (LLM) specialized in Japanese document processing. The new version, “PolySphere-3,” significantly outperforms its predecessor, “PolySphere-2,” achieving world-leading accuracy in structuring data.
This update is the result of a research project selected for GENIAC (Generative AI Accelerator Challenge), a Japanese national initiative led by the Ministry of Economy, Trade and Industry and the New Energy and Industrial Technology Development Organization, aimed at strengthening domestic generative AI development. The project, titled “Innovation and autonomy enhancement of semi-structured forms utilizing generative AI infrastructure,” served as the foundation for this advancement.
Research results from the GENIAC project
In the second cycle of GENIAC, AI inside established two models through the advancement of PolySphere-3: “PolySphere-3,” with the world’s highest performance LLM in data structuring accuracy, and the lightweight and fast-processing “PolySphere-3 Lite.”
- 95%+ OCR accuracy
“PolySphere-3” achieved an average OCR accuracy of 95.1% across 50 types of forms, outperforming other major LLMs*1. - Improved processing speed by reducing model weight
“PolySphere-3 Lite” significantly improves processing speed while maintaining comparable OCR accuracy to “PolySphere-2.” - Technology for continuous accuracy improvement
The new model includes a “self-distillation” mechanism that enables autonomous learning and optimization of forms, ensuring continuous accuracy improvement without manual intervention.
*1 Based on AI inside’s proprietary evaluation criteria. OCR accuracy was compared against multiple major LLMs. These results do not constitute a performance guarantee.
Benchmark Repository:
[ https://github.com/aiinside/japanese-bizform-table-kie ]
Social implementation in “DX Suite:” High-performance AI Available for everyone
AI inside has implemented “PolySphere-3” into “DX Suite,” our AI Agent that automates data entry tasks, for improved reading accuracy for semi-structured forms. With this update, all users can benefit from enhanced accuracy in handling forms without additional settings or costs.
- 95%+ OCR accuracy for semi-structured Forms
Internal verification of the top 90% of forms processed via DX Suite confirmed over 95% OCR accuracy, enabling general-purpose, high-accuracy form automation. - Optimal model selection
“DX Suite” has “PolySphere-3” with prioritized accuracy applied by default. Users who require faster processing can opt for the lightweight “PolySphere-3 Lite” model, providing flexibility based on operational needs. - Continuous learning and accuracy improvement
With more than 200 million OCR operations processed monthly, the “self-distillation” mechanism of “DX Suite” resulting from the research ensures continuous improvement in OCR accuracy.
About “DX Suite”
“DX Suite” is an AI Agent designed to automate the entire workflow surrounding data entry tasks. Combining industry-leading OCR accuracy with in-house developed character recognition AI (AI-OCR engine) and advanced technologies specializing in data structuring, AI automates cumbersome work processes that previously relied on manual labor. Analog data from various document formats is digitized with high accuracy and efficiency, then seamlessly converted into usable formats and integrated into downstream business systems.This enables organizations across all industries to significantly improve operational efficiency and data management, enhancing overall productivity and accelerating digital transformation.
Website: https://inside.ai/en/dx-suite
About AI inside Inc.
AI inside is a tech company engaged in the research, development, and social implementation of generative AI, large language models (LLMs), and autonomous AI. We have developed the Japanese-language-optimized large language model “PolySphere” and have delivered our solutions to over 3,000 organizations and 60,000 users, including government agencies, local municipalities, and private enterprises, while continuing to advance the development and adoption of our proprietary AI infrastructure.
Website: https://inside.ai/en
*The service names appearing on this site are trademarks or registered trademarks of each company.
AI inside Inc. (https://inside.ai/en/) Public Relations Unit |