Khalifa University, GSMA and AMD Open RF-GPT Wireless AI Model to Telecom Operators

Operators will get access through an application programming interface for a set evaluation period, with the model running on AMD's computing infrastructure, said Louis Powell, director of AI technologies at the GSMA. He said operators need chances to test new AI capabilities against specific use cases, and that feedback from participants would show where the model adds value and what it needs to develop further.

The project takes RF-GPT out of the lab and puts it in front of the companies that would use it. Most AI tools for telecoms work on text and structured network data. The radio layer itself, where signals compete for spectrum and interfere with one another, has largely been left to specialist engineers and dedicated measurement tools. If operators find the model useful, it would give a university research team in the UAE a route to commercial users through the mobile industry's main trade body.

RF-GPT was developed by the Digital Future Institute at Khalifa University. According to WAM, users can ask it about wireless signals it has observed, identify their characteristics and analyse complex scenes in which several signals overlap. The partners say it could help operators optimise their radio access networks, plan networks, manage spectrum and analyse interference.

The university first unveiled the model in April. At the time it said RF-GPT turns raw radio waveforms into time-frequency spectrograms, which a visual encoder processes before a language model answers questions about them. It said the model was trained on about 625,000 computer-generated radio signal examples and beat baseline models by up to 75.4% on spectrogram tasks. Those results come from the university's own announcement and have not been independently benchmarked. Researchers from Université de Lorraine and Zhejiang University worked on the project with Khalifa University.

Merouane Debbah, senior executive director of the Digital Future Institute, said foundation models give the industry a new way to understand the wireless environment by combining signal understanding with reasoning in natural language. He said the evaluation was meant to let operators test that approach in practice and help shape the model's development around the problems they face in their networks.

Philippe Guido, AMD's executive vice president and chief commercial officer, said the collaboration would let operators judge whether AI can simplify radio-frequency analysis and support network planning, spectrum management and interference reduction.

This is not the first project between the university and the GSMA. At MWC Doha a year earlier, GSMA Foundry, the association's innovation arm, signed an agreement with Khalifa University to develop telecom-specific AI models, datasets and benchmarks. Its first output was a new version of TelecomGPT, a language model for the telecoms industry that the university's 6G Research Centre first built in 2024.

The announcement did not say how long the evaluation period will last, which operators have signed up or whether access will be free. It also gave no timetable for a commercial version of RF-GPT.