How Fujitsu Built Takane, a Japanese LLM for Regulated Industries, with Cohere

Fujitsu serves customers across more than 100 countries as a digital transformation partner, with deep roots in sectors that demand the highest standards for data privacy and regulatory compliance: finance, government, healthcare, law, and manufacturing. As generative AI became a core transformation enabler, Fujitsu recognized that general-purpose LLMs — optimized primarily for English — created a fundamental accuracy gap for Japanese enterprise customers. In regulated industries, even minor language errors can have serious consequences, making reliability in Japanese-specific contexts non-negotiable.

The challenge Fujitsu faced was not simply building a Japanese-capable model, but building one secure enough for industries operating under strict data governance requirements. General LLMs hosted by public cloud providers could not meet these customers’ compliance and sovereignty requirements. Developing a robust, scalable model from scratch would also require access to advanced training infrastructure, extensive Japanese language datasets, and deep expertise in LLM development — capabilities Fujitsu needed to acquire or partner for.

Fujitsu chose Cohere as its strategic technology partner, leveraging the Cohere Command series as the base architecture. Working with Cohere’s engineers, Fujitsu built Takane as a privately-deployed custom model optimized for the specific linguistic and domain requirements of Japanese enterprise customers. The model’s capabilities span complex document workflow processing, multilingual support (Japanese, German, Chinese, Portuguese), structured data extraction into formats like JSON and CSV, mathematical and logical reasoning for finance and engineering use cases, and high-accuracy summarization and sentiment analysis for enterprise decision-making.

The performance outcome was significant: Takane achieved a world-class score on the JGLUE benchmark, validating its accuracy for the demanding business contexts Fujitsu’s customers operate in. The model is now embedded in Fujitsu’s commercial AI infrastructure — integrated into the Fujitsu Kozuchi AI service and offered through the Fujitsu Data Intelligence PaaS (DI PaaS) — making it accessible to customers across Fujitsu’s client base in Japan’s most privacy-sensitive industries.

The Takane partnership demonstrates how an established global IT company can accelerate AI capability development by combining its domain expertise and distribution reach with a specialized AI partner’s model infrastructure. Rather than building its own LLM from zero, Fujitsu used Cohere’s training expertise and base models to reach production quality faster than an independent effort would have allowed, while retaining the ability to deploy privately and maintain full control over its customers’ data.