Generative artificial intelligence has advanced rapidly, but it still struggles when venturing into the more technical territory of mobile networks. Interpreting 3GPP specifications, diagnosing complex incidents, or navigating scattered documentation remains a challenge for many generalist models. In this context, the announcement made at MWC Doha is significant: GSMA Foundry and Khalifa University of Science and Technology have formed a strategic partnership to create TelecomGPT, a large language model (LLM) specialized in the telecom domain, along with new open assets for the industry.
The initiative aims for a very specific goal: to equip operators and suppliers with an AI “brain” that truly understands the telecommunications ecosystem, with guarantees of reliability, security, and energy efficiency.
A strategic agreement born at MWC Doha
The announcement was made in Doha during the first day of MWC Doha and represents a new step in GSMA Foundry’s strategy, the innovation hub of the global mobile operators’ association. Meanwhile, Khalifa University, one of the top higher education institutions in the United Arab Emirates, contributes the research capabilities of its 6G Research Centre (6GRC), which specializes in next-generation technologies.
The partnership focuses on developing:
- AI datasets specific to telecommunications.
- Models and benchmarking frameworks to evaluate their performance on real sector tasks.
- Open, robust, secure, and energy-efficient solutions for the industry.
The collaboration is supported by an explicit commitment to knowledge exchange and to driving the digital transformation of the mobile ecosystem.
TelecomGPT and a telco knowledge graph as initial deliverables
In the first phase, GSMA and Khalifa University will provide the industry with two key open assets:
- TelecomGPT, a “telco-first” LLM designed from the ground up to handle terminology, processes, and use cases specific to the sector. It will feature a chat interface hosted on LightOn’s AI platform.
- A Open Telco Knowledge Graph focused on 3GPP documentation, built on LightOn’s computing infrastructure and hosted on Hugging Face.
The goal is for these resources to facilitate training, evaluation, and deployment of AI models that are truly aligned with the needs of operators, network equipment manufacturers, integrators, and regulators. Combining a specialized LLM with a structured knowledge graph about standards aims to reduce friction when consulting, cross-referencing, and applying complex technical information.
When generalist models fall short: the telco AI gap
GSMA has been working on the Open-Telco LLM Benchmarks, a series of public tests designed to evaluate how large models respond to telecom-specific tasks. The conclusions are clear: even state-of-the-art models face notable difficulties when asked to:
- Accurately interpret technical specifications and standards such as 3GPP.
- Assist in network troubleshooting.
- Handle deep technical knowledge about architectures, interfaces, and protocols.
In other words, generalist models are excellent as general-purpose assistants but not as reliable decision-makers in critical environments where a nuance in a specification can mean the difference between a stable network and a service outage.
TelecomGPT is created specifically to bridge that gap, supported by curated data and knowledge structures tailored to the telco domain.
The role of the 6G Research Centre and a long-term vision
Khalifa University, through its 6G Research Centre, has established itself as a key player in both local and international telecommunications projects. The centre is exploring technologies beyond 5G that anticipate the requirements of future 6G networks: higher device density, new frequency bands, critical services, and tighter integration of networks and edge computing.
In this context, the collaboration with GSMA connects cutting-edge research with concrete industry needs: more reliable models, transparent evaluation frameworks, and open tools that avoid being trapped in proprietary silos.
The university emphasizes that combining resources—such as infrastructure, AI expertise, and telco knowledge—should lead to solutions capable of addressing critical limitations of current models, especially in tasks where technical accuracy is non-negotiable.
Open assets to accelerate telco AI adoption
Both TelecomGPT and the Open Telco Knowledge Graph are conceived as open assets aimed at speeding up AI adoption in the telecom industry. The goal is not to replace internal developments by operators but to provide a solid foundation for building:
- Internal assistants for network engineers and system architects.
- Support tools for operations that understand alarms, events, and standard procedures.
- Decision support systems capable of “reading” and contextualizing 3GPP documentation.
- New use cases in automation, network assurance, or service design.
Being open and shared, these resources also aim to reduce duplication: instead of each company training its own model from scratch, they can leverage a common specialized core and adapt it to their needs.
Specialized AI for a safer and more efficient ecosystem
Beyond operational efficiency, this initiative addresses a broader debate: how to make AI in telecommunications reliable, secure, and sustainable. GSMA and Khalifa University emphasize the importance of developing models that:
- Robust, capable of behaving predictably even with ambiguous or malformed inputs.
- Secure, minimizing risks of incorrect responses that could affect network operation.
- Energy-efficient, a key aspect amid increasing AI compute consumption.
Using common benchmarking frameworks along with assets like TelecomGPT and the knowledge graph should help operators and manufacturers better assess which models are truly prepared to coexist with critical systems.
FAQ about TelecomGPT and the GSMA–Khalifa University collaboration
What is TelecomGPT and what is it designed for?
TelecomGPT is a large language model (LLM) specialized in telecommunications. It is designed to handle technical documentation, standards like 3GPP, and typical sector tasks, from interpreting specifications to assisting in network troubleshooting, offering responses more aligned with the telco context than a generalist model.
How does TelecomGPT differ from other general-purpose generative AI models?
While large generalist models are trained on broad and heterogeneous data, TelecomGPT is built from sector-specific datasets and evaluated with telco benchmarks. This enables it to perform better on tasks like understanding technical standards, handling network acronyms, and following industry procedures—areas where generalist models often fall short.
What is the Open Telco Knowledge Graph based on 3GPP documentation?
It’s a knowledge graph that organizes and relates information from official 3GPP documentation, the organization responsible for mobile network standards. Structuring these contents as a graph makes it easier for models like TelecomGPT to navigate, consult, and cross complex information, improving the quality of technical responses.
How can operators and companies access TelecomGPT and these open assets?
According to the announcement, TelecomGPT will feature a chat interface hosted on LightOn’s AI platform, while the Open Telco Knowledge Graph will be published on Hugging Face. GSMA and Khalifa University aim for these resources to be accessible to the entire mobile ecosystem—operators, device makers, integrators, and research centers—for their AI development and testing in telecommunications.
via: prnewswire