SoftBank’s Large Telecom Model Achieves Top Ranking in AI Benchmarks

In an exciting development for the telecommunications sector, SoftBank recently announced that its generative AI foundation model, the Large Telecom Model (LTM), has secured a prestigious ranking in the GSMA Open-Telco LLM Benchmarks. This accolade recognizes LTM as one of the most effective large language models (LLMs) tailored specifically for telecommunications. The ranking is significant as it evaluates LLMs based on their performance across a variety of telecom-related tasks, marking a leap forward in the intersection of AI and telecom.

SoftBank’s commitment to enhancing LTM doesn’t stop with this recognition. The company has established a telecom-specialized learning framework. This framework capitalizes on the vast data assets and operational expertise gained from SoftBank’s longstanding experience as a telecommunications operator. As a result, the top-tier recognition aligns seamlessly with SoftBank’s mission to improve telecom-domain performance continually through innovative training processes.

As of March 30, 2026, this ranking is based on the average scores from all evaluation datasets among 84 models submitted to the GSMA Open-Telco LLM Benchmarks.

Central to this initiative is the Open Telco AI initiative launched at the MWC Barcelona 2026, aimed at creating a collaborative environment for telco-grade AI solutions. The GSMA brings together telecom operators, vendors, developers, and academia to establish a shared portal for models, datasets, compute, and tools. Within this framework, the Open-Telco LLM Benchmarks serve as a critical component, providing transparency and a means to measure and enhance model performance in real-world telecom scenarios. This focus on accuracy and reliability is vital for the operational needs of modern telecom networks.

The Importance of a Specialized Learning Framework

To achieve the necessary performance for real-world applications, general-purpose LLMs alone are inadequate in the telecommunications domain. These models must grasp not only complex technical standards but also excel in domain-specific question answering and interpreting operational logs. This complexity necessitates a robust learning framework tailored specifically for telecommunications.

The framework set up by SoftBank organizes both data design and training processes to address the intricate structures and interdependencies inherent in telecom network data. By employing a variety of public and proprietary telecommunications datasets—which include network data and insights into network design and management—SoftBank’s LTM can develop a comprehensive understanding of the field.

The Role of Advanced Data Training Techniques

SoftBank’s training approach for LTM involves a multi-faceted strategy. This includes continual pre-training, fine-tuning, and reinforcement learning. Such comprehensive training is essential to enhance LTM’s performance over time. The telecommunications domain consists not only of traditional text-based documents but also features diverse formats like tabular data and code descriptions, making the organization and transformation of these datasets into optimized synthetic data a critical step in the training process.

Moreover, using LLM-based data filtering allows SoftBank to maintain high training data quality. The incorporation of hyperparameter optimization (HPO) utilizing small language models (SLMs) also serves to boost learning efficiency, culminating in an impressive model performance.

Recognition and Future Aspirations

The exceptional advancements achieved through this structured framework have enabled LTM to attain a leading position in the GSMA Open-Telco LLM Benchmarks. The evaluation process scrutinizes model performance against real-world telecom challenges, assessing capabilities such as understanding telecom specifications, answering specific domain inquiries, interpreting operational logs, and performing mathematical reasoning within telecom contexts.

SoftBank’s top-tier ranking showcases not only the effectiveness of LTM but also affirms the company’s substantial contributions to developing generative AI models specifically catered to the telecommunications landscape.

Ryuji Wakikawa, Vice President, Head of the Research Institute of Advanced Technology at SoftBank

“SoftBank has been developing a telecom industry–specific LTM, trained on telecom expertise and real operational data, and achieved top-class results in the GSMA Open-Telco LLM Benchmarks. This demonstrates that our training foundation is also at a high international standard. By leveraging LTM, SoftBank will thoroughly enhance its operations and lead the advancement of the telecommunications industry.”

Louis Powell, Director of AI initiatives, GSMA

“Telecom networks demand precision and context that general-purpose AI often struggles to deliver. By testing models against telecom-relevant datasets and tasks, the GSMA Open-Telco LLM Benchmarks spotlight genuine capability improvements. SoftBank’s top-tier ranking is a strong example of that progress, and exactly the kind of momentum the industry needs as it scales AI responsibly into operations.”