Launch of PLaMo Translate for Use in the Government AI Project "Gennai"

Overview of the Launch

The Digital Agency has decided to provide government employees with access to PLaMo Translate, developed by Preferred Networks, Inc. (PFN), under the generative AI environment (project named “Gennai”).
Internal use within the Digital Agency will launch in December, with a rollout to other ministries and agencies planned from 2026 onward.

Advancing the Use of Generative AI in Government

In order to advance Government AI, the Digital Agency is currently deploying the project Gennai across ministries and agencies. This initiative aims to create an environment where government employees can safely utilize generative AI in their daily work, while ensuring proper security for information systems.

In particular, the use of large language models (LLMs) developed and provided by domestic companies, such as PFN's “PLaMo Translation,” optimized for specific Japanese expressions and writing styles found in government documents (hereinafter referred to as “domestically developed AI system”), is essential for promoting the safe and secure use of generative AI while ensuring the reliability of government operations.

Through collaboration with domestic companies and related entities, the Digital Agency will advance the development of an environment in which domestically developed AI systems, trained on high-quality dataset containing Japanese culture and practices and rich in Japanese language, can also be actively utilized, with the aim of both generating effective use cases of domestically developed AI systems in administrative operations and establishing a continuous cycle for improving their accuracy.

(Reference) Outline and Features of “PLaMo Translate”

This domestically developed large language model is entirely designed in Japan, from its architecture to training, and specializes in Japanese-to-English and English-to-Japanese translation. Unlike models based on existing overseas systems, it generates fluent and natural Japanese translations with minimal repetition, omissions, or inconsistent wording, even for long texts. Optimized for a variety of styles and contexts, including conversational text, news articles, and academic papers, it delivers coherent and natural translations. The model can also effectively handle the specific vocabulary and phrasing often found in government documents, ensuring accurate and readable output across all text types.