TOKYO, Sept. 8, 2025 /PRNewswire/ — Fujitsu announced the development of a new reconstruction technology for generative AI. The new technology, positioned as a core component of the Fujitsu Kozuchi AI service, will strengthen the Fujitsu Takane LLM by enabling the creation of lightweight, power-efficient AI models.
The advancement is underpinned by two key innovations: quantization and specialized AI distillation. Fujitsu’s proprietary 1-bit quantization technology applied to Takane achieves a remarkable 94% reduction in memory consumption, while maintaining an unprecedented 89% accuracy retention rate compared to unquantized models. This leads to a three-fold increase in inference speed, significantly outperforming conventional quantization methods. Consequently, large generative AI models previously requiring multiple high-end GPUs can now efficiently operate on a single low-end GPU.
Fujitsu’s world-first specialized AI distillation not only significantly reduces model size, but also enhances accuracy beyond that of the original model. This brain-inspired approach extracts and condenses task-specific knowledge, creating highly efficient and reliable specialized AIs.
This revolutionary lightweighting capability promises to democratize cutting-edge AI, enabling the deployment of sophisticated agentic AI on edge devices such as smartphones and factory machinery. This will lead to improved real-time responsiveness, enhanced data security, and a radical reduction in power consumption for AI operations, significantly contributing to a sustainable AI society.
Fujitsu plans to roll out trial environments for Takane with this applied technology starting in the second half of fiscal year 2025 and will progressively release models of Cohere’s research open-weight Command A quantized via Hugging Face. Fujitsu remains committed to advancing generative AI capabilities to solve complex societal challenges and unlock new possibilities for AI utilization.
For full release click here
