SoftBank new GenAI model LTM uses homegrown Sarashina LLM to optimize network operations securely within Japan.
SoftBank Corp. has launched a Japan-developed GenAI foundation model called the Large Telecom Model (LTM). This integrates its proprietary Sarashina LLM to power telecom operations. This upgrade enables end-to-end AI training and inference within Japan, addressing critical data privacy and localization challenges.
Previously, SoftBank relied on general-purpose models trained on English data, which failed to grasp Japan’s complex language and industry-specific nuances. The integration of Sarashina changes that. It allows the LTM to understand technical terminology, contextual cues, and cultural references unique to Japan. This creates a fully domestic, secure AI ecosystem for telecom operations.
SoftBank is already using the GenAI-powered LTM internally to automate mobile network management, a process that once required days but now takes minutes. In tests during Tokyo’s Kita City Fireworks Festival, the model predicted communication quality with over 90% accuracy, helping engineers preempt connectivity issues during large-scale events. By automating base station configuration, LTM reduces human error and dependency on expert oversight, improving efficiency and service reliability.
To enhance its intelligence further, SoftBank has embedded reinforcement learning and reasoning capabilities into the model. These upgrades allow LTM to learn autonomously from radio wave simulations, optimize network conditions, and explain its decisions through visualized reasoning paths. The company plans to expand LTM’s use across operations and collaborate with research partners to refine its models.
SoftBank’s GenAI-led approach marks a shift from automation to autonomy in telecom management. By combining homegrown AI infrastructure with real-time optimization, it’s creating a secure, scalable, and self-improving network intelligence framework that could define the future of digital connectivity in Japan.
