Telecom operators can now test an AI model that reads radio signals and answers questions about them in plain language. Khalifa University of Science and Technology, the GSM Association (GSMA) and chipmaker AMD have opened an operator trial for RFGPT, a radio frequency foundation model built at the university’s Digital Future Institute. The trial launched at MWC26 Doha, which opened in Doha today.
Operators who join get two weeks of access to the model through an application programming interface, or API. They apply by sending in a use case, and the research team then agrees the scope of the test, the access setup and the dates with each operator.
An AI model that reads radio signals like text
RFGPT is the first radio frequency language model that turns radio signals into spectrograms, which are pictures of how signal energy changes over time and frequency, and then reasons about them in words. That means an engineer can ask what a signal is, instead of reading it through specialist tools.
The model itself was first announced in February 2026, so the trial is the new step, not the model. It was trained on around 625,000 computer made examples drawn from about 12,000 simulated scenes across six wireless technologies, with no manual labelling needed.
In benchmark tests, RFGPT beat OpenAI’s GPT-4 and Alibaba’s Qwen 2.5 on tasks like working out modulation types and pulling out 5G information. It also counted the number of signals in a spectrogram correctly about 98 percent of the time, a task general purpose models almost never get right.
What operators can test in the two week trial
The trial covers the everyday jobs that keep mobile networks running. Operators can test the model for network planning, spectrum management, interference analysis and radio access network optimisation.
RFGPT can spot wireless technologies including Wi-Fi, Bluetooth, LTE, 5G New Radio, UMTS and DVB-S2, and tell uplink from downlink where that applies. It can also read modulation schemes, pick apart crowded spectrum with several signals at once, and work out where transmissions overlap in time and frequency.
It can even estimate how busy a network is, by counting Wi-Fi and 5G users in a given window, and it can read key 5G settings such as subcarrier spacing and synchronisation signal block patterns. Each operator will be asked to report back on how well the model performed and how useful it was for their own work.
From a February launch to a live trial in Doha
The trial builds on a partnership that started a year ago. GSMA Foundry, the innovation arm of the GSMA, and Khalifa University began working together in November 2025, with the aim of building open telecom AI tools and benchmarks.
The work is led by Professor Merouane Debbah, senior director of the university’s Digital Future Institute, with a team from Khalifa University, Universite de Lorraine and Zhejiang University. The model runs on AMD computing hardware, so the trial also gives operators a chance to test AMD’s AI infrastructure.
MWC26 Doha runs from 8 to 10 November at the Doha Exhibition and Convention Center, under the theme The IQ Era. It sits alongside the International Telecommunication Union’s Plenipotentiary Conference, which runs at the same venue from 9 to 27 November.
What RFGPT means for the UAE’s telecom future
The most interesting thing about this trial is how practical it is. A lot of AI news is about bigger models and bigger promises, but RFGPT is about a concrete job: helping the people who run networks understand what is happening in the airwaves, faster and in plain words.
If it works, the payoff is real. Networks get more crowded every year, and the UAE is pushing hard into 6G research. Tools that let engineers ask questions instead of decoding charts could cut the time it takes to fix faults and plan capacity. The trial will show whether that promise holds up in live networks, and that test matters more than any benchmark score.
