SoftBank Open-Sources AITRAS Dynamic Scoring Framework for AI-RAN

Tokyo, Japan — SoftBank Corp. announced Wednesday the open-source release of a core component of its artificial intelligence network orchestration system.

The company stated the release is intended to facilitate broader industry adoption of AI-integrated mobile network infrastructure.

The released component, called the Dynamic Scoring Framework, is part of SoftBank's AITRAS orchestrator, a system designed to manage computing resources, including graphics processors, central processors, and AI accelerators, across multiple locations and hardware clusters.

The framework evaluates and ranks available resources according to metrics such as power consumption and processing efficiency.

The system then automatically allocates workloads to the most suitable hardware.

The release is intended to advance AI-RAN, a technical approach that enables AI applications and conventional mobile network functions to operate concurrently on shared infrastructure.

SoftBank contributed the framework to Open Cluster Management, an open-source project that coordinates multiple Kubernetes clusters.

The company stated that making orchestration tools publicly available could enable telecommunications carriers to allocate development resources to network-specific services.

The project was developed in collaboration with Red Hat. 

SoftBank is working to integrate the framework with Red Hat's OpenShift and Advanced Cluster Management platforms.

The companies have also been testing llm-d, an open-source tool designed to optimize AI inference workloads.

According to SoftBank, tests indicate the system can automatically assign processing roles to specific chips based on real-time resource availability, resulting in improved performance compared with manual configuration.

SoftBank plans to demonstrate the technologies at the Red Hat booth during MWC Barcelona 2026.