Even as the world of IT and communications was coming to terms with the potential ramification of the agentic artificial intelligence (AI)-based hack on AI company Hugging Face, US comms giant AT&T has launched OTel 2.0, a dedicated model for the telecoms industry.
The telco says the announcement highlights a broader shift towards domain-specific AI. That is rather than relying solely on frontier models, it is working with global mobile trade association the GSMA, Microsoft, AMD, Dell and Red Hat to build specialised telecom models that improve accuracy for network operations while reducing infrastructure costs.
The offer is built on Gemma 4 31B-IT, an open multimodal model built by Google DeepMind that handles text and image inputs and can process video as sequences of frames, and generates text output. It was trained using 400 billion telecom-specific tokens selected from more than 1 trillion processed tokens.
Alongside the model, AT&T has revealed it has developed an AI Gateway that intelligently routes prompts to the most cost-effective model for each task. Processing an average of 45 billion AI tokens every day, the system is already claimed to be cutting AI inference costs by up to 90%, saving the company millions while maintaining performance.
Explaining the reasons for the launch, AT&T noted that artificial intelligence was now a cornerstone of telecommunications’ future, and that because it worked with what it called some of the most complex data sets in the world, it had a real opportunity to transform how to automate complex network operations and deliver more personalised customer experiences.
However, stressed AT&T chief data and AI officer Andy Markus, the company could not lose sight of one of the most critical responsibilities: using AI efficiently. A new, intelligent AI gateway could crack the cost challenge that is rapidly emerging with AI usage.
“We’re focused on efficiency in two primary ways: we’re using a proprietary cache-aware router to select the most cost-effective models for each task, and we’re training more open-source models to address telco-specific needs for accuracy,” he said. “And both of these approaches are being actively used in production to drive real results, not just theoretical ones.
“Running advanced AI models can be expensive, especially at AT&T’s scale: an average of 45 billion tokens per day. People often default to the latest and greatest models, but only a small percentage of the tasks we run require that level of sophistication. Many can be handled by lower-cost models without sacrificing performance … We’ve built an AI Gateway that goes beyond simple model selection: it uses cache-aware routing to intelligently match each task to the most cost-effective model – without compromising on quality. At each turn, the gateway weighs speed and cost with the expected quality of the output, then routes the prompt to the best model.”
For its part, the GSMA stressed how the telecoms industry simply needs its own models. To that end, it said general-purpose AI models have come a long way, but they weren’t built with telecoms in mind. It believes that to ask a general-purpose AI model to interpret an industry standard or troubleshoot a live network issue, “the cracks start to show”, not because the models are not capable, but because the data they learned from “barely touches this domain”.
In a blog post, it added: “That gap shows up in the results. The top three performers on the Open Telco AI benchmarks are all domain-adapted models, not general-purpose ones, with the new OTel 2.0 model top of the Open Telco AI leaderboard. This clearly illustrates that domain adapted models are highly accurate and can be significantly smaller in size”
Moreover, the GSMA noted that just as healthcare, financial services and manufacturing are developing domain-specific AI approaches, the telecoms industry needs models trained on its unique standards, protocols and operating environments.
This, it said, was not just about accuracy, it was about enterprise requirements; reducing costs and maintaining control by deploying models across clouds and on-premise as needed. The GSMA concluded that accuracy and requirements together are vital for operators deploying telco-specific use cases, like network troubleshooting, product development, network configuration and more.
