Multi-Modal Intelligent Channel Modeling: From Fine-tuned LLMs to Pre-trained Foundation Models

To meet the evolving demands of sixth-generation (6G) wireless channel modeling, such as precise prediction capability, extension capabilities, and system participation capability, multi-modal intelligent channel modeling (MMICM) has been proposed based on Synesthesia of Machines (SoM) which explores the mapping relationship between multi-modal sensing in physical environment and channel characteristics in electromagnetic space. Furthermore, for integrating heterogeneous sensing, reasoning across scales, and generalizing to complex air-space-ground-sea communication environments, two new paradigms of MMICM are explored, including fine-tuned large language models (LLMs) for Channel Modeling (LLM4CM) and Wireless Channel Foundation Model (WiCo). LLM4CM leverages pre-trained LLMs on channel representations for cross-modal alignment and lightweight adaptation, enabling flexible channel modeling for 6G multi-band and multi-scenario communication systems. WiCo, which pre-trained on physically valid channel realizations and their associated environmental and modal observations, embeds electromagnetic equations for physical interpretability and uses parameterized adapters for scalability. This article details the architectures and features of LLM4CM and WiCo, laying a foundation for artificial intelligence (AI)-native 6G wireless communication systems. Then, we conducts a comparative analysis of the two emerging paradigms, focusing on their distinct characteristics, relative advantages, inherent limitations, and performance attributes. Finally, we discuss the future research directions.

Bootstrapping Exploration with Group-Level Natural Language Feedback in Reinforcement Learning

Large language models (LLMs) typically receive diverse natural language (NL) feedback through interaction with the environment. However, current reinforcement learning (RL) algorithms rely solely on scalar rewards, leaving the rich information in NL feedback underutilized and leading to inefficient exploration. In this work, we propose GOLF, an RL framework that explicitly exploits group-level language feedback to guide targeted exploration through actionable refinements. GOLF aggregates two complementary feedback sources: (i) external critiques that pinpoint errors or propose targeted fixes, and (ii) intra-group attempts that supply alternative partial ideas and diverse failure patterns. These group-level feedbacks are aggregated to produce high-quality refinements, which are adaptively injected into training as off-policy scaffolds to provide targeted guidance in sparse-reward regions. Meanwhile, GOLF jointly optimizes generation and refinement within a unified RL loop, creating a virtuous cycle that continuously improves both capabilities. Experiments on both verifiable and non-verifiable benchmarks show that GOLF achieves superior performance and exploration efficiency, achieving 2.2$\times$ improvements in sample efficiency compared to RL methods trained solely on scalar rewards. Code is available at https://github.com/LuckyyySTA/GOLF.

From Specialist to Large Models: A Paradigm Evolution Towards Semantic-Aware MIMO

The sixth generation (6G) network is expected to deploy larger multiple-input multiple-output (MIMO) arrays to support massive connectivity, which will increase overhead and latency at the physical layer. Meanwhile, emerging 6G demands such as immersive communications and environmental sensing pose challenges to traditional signal processing. To address these issues, we propose the ``semantic-aware MIMO'' paradigm, which leverages specialist models and large models to perceive, utilize, and fuse the inherent semantics of channels and sources for improved performance. Moreover, for representative MIMO physical-layer tasks, e.g., random access activity detection, channel feedback, and precoding, we design specialist models that exploit channel and source semantics for better performance. Additionally, in view of the more diversified functions of 6G MIMO, we further explore large models as a scalable solution for multi-task semantic-aware MIMO and review recent advances along with their advantages and limitations. Finally, we discuss the challenges, insights, and prospects of the evolution of specialist models and large models empowered semantic-aware MIMO paradigms.

Variational Trajectory Optimization of Anisotropic Diffusion Schedules

We introduce a variational framework for diffusion models with anisotropic noise schedules parameterized by a matrix-valued path $M_t(θ)$ that allocates noise across subspaces. Central to our framework is a trajectory-level objective that jointly trains the score network and learns $M_t(θ)$, which encompasses general parameterization classes of matrix-valued noise schedules. We further derive an estimator for the derivative with respect to $θ$ of the score that enables efficient optimization of the $M_t(θ)$ schedule. For inference, we develop an efficiently-implementable reverse-ODE solver that is an anisotropic generalization of the second-order Heun discretization algorithm. Across CIFAR-10, AFHQv2, FFHQ, and ImageNet-64, our method consistently improves upon the baseline EDM model in all NFE regimes. Code is available at https://github.com/lizeyu090312/anisotropic-diffusion-paper.

VESPO: Variational Sequence-Level Soft Policy Optimization for Stable Off-Policy LLM Training

Training stability remains a central challenge in reinforcement learning (RL) for large language models (LLMs). Policy staleness, asynchronous training, and mismatches between training and inference engines all cause the behavior policy to diverge from the current policy, risking training collapse. Importance sampling provides a principled correction for this distribution shift but suffers from high variance; existing remedies such as token-level clipping and sequence-level normalization lack a unified theoretical foundation. We propose Variational sEquence-level Soft Policy Optimization (VESPO). By incorporating variance reduction into a variational formulation over proposal distributions, VESPO derives a closed-form reshaping kernel that operates directly on sequence-level importance weights without length normalization. Experiments on mathematical reasoning benchmarks show that VESPO maintains stable training under staleness ratios up to 64x and fully asynchronous execution, and delivers consistent gains across both dense and Mixture-of-Experts models. Code is available at https://github.com/FloyedShen/VESPO

Large Wireless Foundation Models: Stronger over Bigger

AI-communication integration is widely regarded as a core enabling technology for 6G. Most existing AI-based physical-layer designs rely on task-specific models that are separately tailored to individual modules, resulting in poor generalization. In contrast, communication systems are inherently general-purpose and should support broad applicability and robustness across diverse scenarios. Foundation models offer a promising solution through strong reasoning and generalization, yet wireless-system constraints hinder a direct transfer of large language model (LLM)-style success to the wireless domain. Therefore, we introduce the concept of large wireless foundation models (LWFMs) and present a novel framework for empowering the physical layer with foundation models under wireless constraints. Specifically, we propose two paradigms for realizing LWFMs, including leveraging existing general-purpose foundation models and building novel wireless foundation models. Based on recent progress, we distill two roadmaps for each paradigm and formulate design principles under wireless constraints. We further provide case studies of LWFM-empowered wireless systems to intuitively validate their advantages. Finally, we characterize the notion of "large" in LWFMs through a multidimensional analysis of existing work and outline promising directions for future research.

WiFo-E: A Scalable Wireless Foundation Model for End-to-End FDD Precoding in Communication Networks

Accurate precoding in massive multiple-input multiple-output (MIMO) frequency-division duplexing (FDD) systems relies on efficient channel state information (CSI) acquisition. End-to-end learning frameworks improve performance by jointly optimizing this process, but they lack scalability and fail to generalize across different system configurations, such as varying numbers of antennas and users. To overcome this limitation, we introduce WiFo-E, a wireless foundation model designed for scalable end-to-end precoding. WiFo-E employs multi-task pretraining on a diverse set of configurations to learn transferable representations of underlying wireless principles. Central to the model is a sparse Mixture-of-Experts (MoE) Transformer architecture, which mitigates task interference and enhances training efficiency by activating specialized parameter subsets adaptively. Extensive simulations demonstrate that WiFo-E outperforms conventional per-configuration training and shows strong generalization to unseen system configurations, providing a flexible and efficient foundation for adaptive massive MIMO precoding.

WiFo-M$^2$: Plug-and-Play Multi-Modal Sensing via Foundation Model to Empower Wireless Communications

The growing adoption of sensor-rich intelligent systems has boosted the use of multi-modal sensing to improve wireless communications. However, traditional methods require extensive manual design of data preprocessing, network architecture, and task-specific fine-tuning, which limits both development scalability and real-world deployment. To address this, we propose WiFo-M$^2$, a foundation model that can be easily plugged into existing deep learning-based transceivers for universal performance gains. To extract generalizable out-of-band (OOB) channel features from multi-modal sensing, we introduce ContraSoM, a contrastive pre-training strategy. Once pre-trained, WiFo-M$^2$ infers future OOB channel features from historical sensor data and strengthens feature robustness via modality-specific data augmentation. Experiments show that WiFo-M$^2$ improves performance across multiple transceiver designs and demonstrates strong generalization to unseen scenarios.

TravelBench: A Broader Real-World Benchmark for Multi-Turn and Tool-Using Travel Planning

Travel planning is a natural real-world task to test large language models (LLMs) planning and tool-use abilities. Although prior work has studied LLM performance on travel planning, existing settings still differ from real-world needs, mainly due to limited domain coverage, insufficient modeling of users' implicit preferences in multi-turn conversations, and a lack of clear evaluation of agents' capability boundaries. To mitigate these gaps, we propose \textbf{TravelBench}, a benchmark for fully real-world travel planning. We collect user queries, user profile and tools from real scenarios, and construct three subtasks-Single-Turn, Multi-Turn, and Unsolvable-to evaluate agent's three core capabilities in real settings: (1) solving problems autonomously, (2) interacting with users over multiple turns to refine requirements, and (3) recognizing the limits of own abilities. To enable stable tool invocation and reproducible evaluation, we cache real tool-call results and build a sandbox environment that integrates ten travel-related tools. Agents can combine these tools to solve most practical travel planning problems, and our systematic verification demonstrates the stability of the proposed benchmark. We further evaluate multiple LLMs on TravelBench and conduct an in-depth analysis of their behaviors and performance. TravelBench provides a practical and reproducible evaluation benchmark to advance research on LLM agents for travel planning.\footnote{Our code and data will be available after internal review.

WiFo-MUD: Wireless Foundation Model for Heterogeneous Multi-User Demodulator

Multi-user signal demodulation is critical to wireless communications, directly impacting transmission reliability and efficiency. However, existing demodulators underperform in generic multi-user environments: classical demodulators struggle to balance accuracy and complexity, while deep learning-based methods lack adaptability under heterogeneous configurations. Although diffusion models have been introduced for demodulation, their flexibility remains limited for practical use. To address these issues, this work proposes WiFo-MUD, a universal diffusion-based foundation model for multi-user demodulation. The model aligns inter-user signal-to-noise ratio imbalance and performs conditional denoising via a customized backbone. Furthermore, a communication-aware consistency distillation method and a dynamic user-grouping strategy are devised to enhance inference. WiFo-MUD achieves state-of-the-art results on large-scale heterogeneous datasets, demonstrating efficient inference and strong generalization across varying system configurations.