Japan’s NTT Opens Access to Latest LLM That Offers Improved Understanding of Japanese Texts

The Yomiuri Shimbun

6:00 JST, October 23, 2025

Japanese telecom company NTT Inc. on Monday began providing access to tsuzumi 2, its latest large language model (LLM). LLMs are a basic component of generative AI.

The latest model has an improved ability to understand Japanese texts, NTT said. The model also has more extensive knowledge in finance, medical care and public administration. According to NTT, tsuzumi 2 is capable of processing Japanese at a level comparable to that of U.S.-based OpenAI’s ChatGPT.

Tsuzumi 2 has four times as many parameters — which indicate the scale of the model’s learning — as the previous model, allowing it to handle more complex processing. NTT expects client companies to use the LLM to create documents, summarize articles and help devise strategies.

According to NTT, tsuzumi 2 can be deployed within a company’s own physical infrastructure, such as its servers. That lets companies avoid storing confidential data on external servers. The telecom company expects that organizations with strict information management requirements, such as financial institutions and local governments, will be interested in the LLM. NTT forecasts LLM orders to exceed ¥500 billion in fiscal 2027, a tenfold increase from current sales.

Companies from countries such as the United States and China are leading in LLM development. Given the situation, NTT is working to develop a model with strengths in particular fields and in processing Japanese.

At a press conference on Monday, NTT President and CEO Akira Shimada said that tsuzumi 2 is a “domestically developed LLM containing NTT’s 40 years of study on the Japanese language.”

“Losing ground to [foreign competitors] is not an option,” he added.