OpenMaxwell

An open research group · Nanyang Technological University

LLMs and agents for wireless.

We use wireless communications as a testbed for large language models and agents: problems that come with derivations, simulators and measurements, so every answer can be checked against physics.

See the projectsJoin the group

4,027
problems in WirelessMathBench-XL
100
ray-traced cities in RadioVISTA
19
models tested on radio fields

Directions

Three questions we work on

01

Reasoning

Can a model derive the equations an engineer would write, and can we tell when it has not?

02

Perception

Can a model infer the radio field a city shapes from what an image or a few measurements show?

03

Agents

How should agents exchange messages and tasks across a network, and what has to be standard?

Projects

Papers, benchmarks and datasets

All projects →

01 ReasoningNeurIPS 2026

WirelessMathBench-XL

An Auditable Benchmark for Wireless Mathematical Reasoning

Xin Li, Mengbing Liu, Yiyang Zhu, Wenhe Zhang, Li Wei, Jiancheng An, Chau Yuen

Rome: the ray-traced radio field (left) and the rendered city (right)

02 PerceptionPreprint 2026

RadioVISTA

Seeing the Invisible: How Vision-Language Models Read, Predict and Trace Radio Fields

Yiyang Zhu, Xin Li, Guangyuan Liu, Ziheng Liu, Jiayi Zhang, Chau Yuen

cities on 6 continents
100
tasks with exact answers
12
models evaluated
19

02 PerceptionPreprint 2026

Radio2Map

Reading 3D Cities from Sparse Radio Measurements with an Image Editor

Xin Li, Yiyang Zhu, Chau Yuen

03 AgentsAI4NextG @ NeurIPS 2025

LACP

LLM Agent Communication Protocol (LACP) Requires Urgent Standardization: A Telecom-Inspired Protocol is Necessary

Xin Li, Mengbing Liu, Chau Yuen

01 ReasoningACL 2025 Findings

WirelessMathBench

A Mathematical Modeling Benchmark for LLMs in Wireless Communications

Xin Li, Mengbing Liu, Li Wei, Jiancheng An, Mérouane Debbah, Chau Yuen

questions
587
source papers
40
best accuracy (DeepSeek-R1)
38.05%

Join us

OpenMaxwell is open to students and researchers who want to build with us.

If you work on language models or on wireless systems and have a benchmark, a dataset or an agent in mind, read how projects run and how authorship works.

How to join