Document Processing · Enterprise Knowledge Search
Fujitsu needed a highly accurate Japanese large language model for customers in regulated industries that handle confidential data and strict compliance requirements, but building that model required advanced infrastructure, large language datasets, and specialized AI expertise it wanted to accelerate through a trusted partner.
Reported impact
Takane is now part of Fujitsu's digital transformation offerings and gives Japanese businesses in regulated industries a secure way to automate document-heavy workflows, improve decision-making, and deliver better customer experiences while Fujitsu accelerates development of high-impact AI applications.
Fujitsu partnered with Cohere to build Takane, a privately deployed Japanese large language model based on the Cohere Command series. Cohere provided training support and technical expertise so Fujitsu could create a model tuned for Japanese enterprise language and capable of world-class performance on the JGLUE benchmark. Takane is designed for high-stakes legal, finance, healthcare, government, and manufacturing workflows where accuracy and security matter, and it processes long text with enough precision to streamline document-heavy operations. The model can extract key details and transform unstructured information into JSON, HTML, Markdown, and CSV, while also supporting multilingual work in languages such as German, Chinese, and Portuguese. Fujitsu now delivers Takane through Fujitsu Kozuchi and its Data Intelligence PaaS so regulated customers can adopt secure generative AI inside broader digital-transformation programs.
Regulated-industry text and business documents enter Takane, the model applies Japanese-centric language understanding and reasoning to process or extract the content, outputs structured formats such as JSON or CSV when needed, and Fujitsu exposes those capabilities through Kozuchi and DI PaaS for downstream enterprise workflows.
Built on the Cohere Command series, and privately deployed
Fujitsu partnered with Cohere to build Takane, a state-of-the-art Japanese LLM for businesses in regulated industries
Takane is also being integrated into Fujitsu’s AI service Fujitsu Kozuchi and offered through the all-in-one operation platform Fujitsu Data Intelligence PaaS (DI PaaS).
Takane processes extended text inputs with precision, streamlining complex, document-heavy workflows
Takane transforms unstructured information into ready-to-use formats like JSON, HTML, Markdown, and CSV.
Review how this reference was scored and trace each normalized claim back to its documented source.
Scoring methodology
Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.
Traceability
Field coverage shows which facts each source supports. Dates are shown only when documented by the source.
Related
OpenAI Sample Oct 8, 2026 Cooley Capital markets involve huge amounts of information that need to be synthesized and analyzed. "For our lawyers, it’s always very busy," explains David Wang, Chief Innovation Officer at Cooley. Cooley is an international law firm with a reputation for helping companies navigate capital markets and… ChatGPT Work · GO Public OpenAI Sample Oct 8, 2026 Gilbert + Tobin Gilbert + Tobin needed to improve research, drafting, analysis, and operational workflows across a law firm without compromising client confidentiality, governance requirements, or human accountability. 87% active usage among enabled ChatGPT users ChatGPT Enterprise · ChatGPT Work · Codex OpenAI Sample Aug 7, 2026 HSP GRUPPE HSP wanted AI to create more capacity across tax advisory, legal research, client communication, and financial analysis without weakening confidentiality, governance, or professional accountability. 40,000+ estimated annual hours of additional capacity AI Client Communication · Booking Assistant SKR03 & SKR04 · ChatGPT Enterprise 
