Noetra, backed by 44 Japanese companies including SoftBank and MUFG Bank, is working on the development of "multimodal foundation models" that serve as a foundation for companies to build customized AI models. Multimodal refers to the ability to handle multiple types of data, such as language, images, audio, and video.
Model ReleasesNoetraHirotora Tanba
Noetra, developing domestic AI foundation models, aims for world-class performance
This article is a translation. Read the Japanese original
Hirotora Tamba, CEO of Noetra, has pointed out the risks associated with the dependence on overseas models in Japan's generative AI usage, such as restrictions on model provision and cloud access. The company aims to ensure "model sovereignty" by placing models on IT infrastructure that Japanese companies can manage themselves.
The foundation models being developed by the company aim to function as a "mothership" encompassing diverse data and tasks. Companies can create models tailored to their needs by extracting data from this mothership or performing additional training with their own data. Furthermore, they envision the creation of small models for specific purposes using a technique called "knowledge distillation," which transfers knowledge from the mothership to smaller models.
The completed models are expected to be provided in an "open-weight" format, where the pre-trained weights (parameters) are released.
Sources
- 国産AIのノエトラ・丹波社長「今からでも1位を取れる」 挑むは“母艦モデル”開発、国内基盤での提供 勝ち筋は （ITmedia AI+、2026-09-29）