Upstage Unveils Japanese-Specialized Model "Shin Pro," Tops Performance Index

Outperforms OpenAI Inference Model in Inference Performance

Upstage announced on the 23rd that it has unveiled "Shin Pro," a large language model (LLM) specialized in Japanese, jointly developed with the Japanese AI specialist company Karakuri.

Shin Pro is a 31-billion-parameter model that has been fine-tuned to precisely reflect Japanese language and cultural context, based on Upstage's proprietary LLM "Solar."

Upstage announced on the 23rd that it has unveiled "Shin Pro," a large language model (LLM) specialized in Japanese, jointly developed with the Japanese AI specialist company Karakuri. Photo by Upstage

This model is designed as an on-premise solution, offering both data security and cost efficiency. It is engineered to allow for the safe use of AI on a company's own infrastructure, and there is reportedly strong demand for on-premise models among Japanese companies.

Shin Pro ranked first with the highest performance among models developed in Japan on the "Weights & Biases Nezumi Leaderboard," the most authoritative Japanese LLM performance index. It also received the highest score among models with 31 billion parameters or fewer, and its inference performance surpassed that of OpenAI's open-weight inference model, "GPT-OSS 120B."

According to Upstage, Shin Pro is currently the only model developed and trained in Japan to rank among the top on the Japanese LLM leaderboard. All other top-ranking models are from the United States and China, including those from OpenAI, Google, Anthropic, and DeepSeek.

Hiroyuki Matsushita, Head of Upstage Japan, stated, "Shin Pro is a meaningful achievement that proves Upstage's technological capabilities can ensure competitiveness even with AI models trained in Japan. Through Shin Pro, which combines industry-specific expertise, security, and an understanding of cultural context, we will actively support the successful adoption of AI by Japanese companies."