Many general-purpose models struggle to interpret network data, understand standard documentation, or automate network operations with sufficient accuracy. This performance gap is limiting progress, with only 16% of communications GenAI deployments1 being applied to network operations.
GSMA today launched Open Telco AI, a global industry initiative aimed at accelerating telco-grade AI through open collaboration across carriers, vendors, AI developers, and academic institutions. This announcement introduces a new portal for Telco Open Models, Data, Computing, and Tools to accelerate the development and evaluation of Telco-focused AI models, accessible at GSMA.com/open-telco-ai. .
Although frontier AI models are rapidly advancing, they continue to underperform on communication-specific tasks. Many general-purpose models struggle to interpret network data, understand standard documentation, or automate network operations with sufficient accuracy. This performance gap limits progress: only 16% of telecom GenAI deployments1 Applied to network operation.
Open Telco AI addresses this challenge by bringing together industry and academic partners to build a foundation of telco-grade AI models, data, compute, benchmarks, and communities. Progress is tracked through a Communication Capability Index that measures the model’s performance across a growing range of communication-specific tasks.
As founding supporters of Open Telco AI, AT&T and AMD are making significant contributions. AT&T releases a family of open communication models developed and trained on open, public data that is hardware and cloud agnostic, demonstrating that AI can deliver value in projects of all sizes and with varying levels of computing resources. AMD provides compute power to train, fine-tune, infer, and evaluate models through its GPU platform, cloud partner TensorWave, and open toolchains.
This effort is also supported by community programs that bring together developers, researchers, and operators to solve real-world communications AI problems. This includes competitions such as the AI Telco Troubleshooting Challenge. We gathered over 1,000 registrations and will announce the winners at MWC26 Barcelona .
“Today’s AI models still do not meet the complexity, accuracy, and reliability that the telecom industry demands. Simply put, AI is not yet speaking for the telecom industry, and carriers are often deploying technologies that fail to meet the required levels of accuracy, safety, and efficiency. Establishing clear benchmarks and collaborating across the industry on datasets, models, and agent systems is essential. Open Telco AI provides a shared foundation designed to bridge this gap, an approach that other regulated sectors such as finance and healthcare can also follow.”
-Lewis Powell, Director of GSMA AI Initiatives
“Telco networks are one of the most demanding and regulated environments for AI, and moving from promising demos to telco-grade performance requires an open foundation for data, workloads, and compute. Through Open Telco AI with GSMA and AT&T, AMD provides the enterprise and AI compute needed to efficiently train, fine-tune, and run open telco-grade models from core to edge.”
– Philippe Guido, Executive Vice President and Chief Commercial Officer, AMD
“The telecom industry needs AI that understands the realities of the network, not just general-purpose models repurposed for telco applications. Through Open Telco AI, AT&T is helping build the datasets, models, and evaluation frameworks that enable telco-grade AI at scale. By providing expertise and creating a realistic test environment, we are helping build generative and agential AI. This collaboration with GSMA is accelerating the industry’s path to intelligent and automated networks. ”
– Andy Markus, AT&T Chief Data and AI Officer
Building an open foundation for telco-grade AI
The new portal supports co-creation of essential building blocks for telco-grade AI. This includes:
- Communication Model: A high-performance open-weight model designed for communication tasks, from network troubleshooting to standards interpretation. This includes AT&T’s models in multiple sizes and architectures, Khalifa University’s Radio Frequency Language Model called RFGPT, and AdaptKey AI’s Large Scale Communication Model (LTM) built on NVIDIA Nemotron.
- Open data: library Knowledge graph, embedding, and fine-tuning datasets of text, logs, and selected standard materials from GSMA, Huawei Technologies France, Khalifa University, Mantis NLP, NetoAI, Pleias, Purdue University, The University of Texas at Dallas, Leeds University, and Yale University, as well as pipeline datasets that generate synthetic data from NVIDIA.
- Compute: Access to the project’s compute and open toolchains to train and infer open models through AMD and TensorWave.
- Benchmark: Leaderboard We evaluate model performance on seven communication-specific benchmarks, along with tools for evaluating and submitting models from your local environment.
- Community: Resources, challenges, and engagement activities to foster collaboration, including the AI Telco Troubleshooting Challenge and Agentic Challenge.
The Open Telco AI initiative is supported by a number of valued contributing partners, including AMD, AT&T, Datumo, Huawei Technologies France, and King Abdullah University of Science and Technology, who submitted data, models, and use cases. KDDI, Khalifa University, KPN, LGU+, Mantis NLP, NetoAI, North Carolina State University, NVIDIA Orange, Ooredoo, Pleias, Purdue University, RelationalAI, SK Telecom, SoftBank, Swisscom, TensorWave, Turkcell, University of Leeds, University of Texas at Dallas, Yale University. Open Telco AI is also supported by valuable participating partners including Adaptive ML, BMC, China Telecom, China Unicom, China Mobile, Deutsche Telekom, DU, e& UAE, Google Cloud, IBM, Liberty Global, Queens University, Telefonica, and Vodafone.
