arXiv subjects

Lihang Pan

Publications and source records attributed to Lihang Pan.

7 recordsLinked to original sources

arXiv2026

GLM-5V-Turbo: Toward a Native Foundation Model for Multimodal Agents

We present GLM-5V-Turbo, a step toward native foundation models for multimodal agents. As foundation models are increasingly deployed in real environments, agentic capability depends not only on language reasoning, but also on the ability to perceive, interpret, and act over heterogeneous contexts such as images, videos, webpages, documents, GUIs. GLM-5V-Turbo is built around this objective: multimodal perception is integrated as a core component of reasoning, planning, tool use, and execution, rather than as an auxiliary interface to a language model. This report summarizes the main improvements behind GLM-5V-Turbo across model design, multimodal training, reinforcement learning, toolchain expansion, and integration with agent frameworks. These developments lead to strong performance in multimodal coding, visual tool use, and framework-based agentic tasks, while preserving competitive text-only coding capability. More importantly, our development process offers practical insights for building multimodal agents, highlighting the central role of multimodal perception, hierarchical optimization, and reliable end-to-end verification.

cs.CV↗

arXiv2025

WebVIA: A Web-based Vision-Language Agentic Framework for Interactive and Verifiable UI-to-Code Generation

User interface (UI) development requires translating design mockups into functional code, a process that remains repetitive and labor-intensive. While recent Vision-Language Models (VLMs) automate UI-to-Code generation, they generate only static HTML/CSS/JavaScript layouts lacking interactivity. To address this, we propose WebVIA, the first agentic framework for interactive UI-to-Code generation and validation. The framework comprises three components: 1) an exploration agent to capture multi-state UI screenshots; 2) a UI2Code model that generates executable interactive code; 3) a validation module that verifies the interactivity. Experiments demonstrate that WebVIA-Agent achieves more stable and accurate UI exploration than general-purpose agents (e.g., Gemini-2.5-Pro). In addition, our fine-tuned WebVIA-UI2Code models exhibit substantial improvements in generating executable and interactive HTML/CSS/JavaScript code, outperforming their base counterparts across both interactive and static UI2Code benchmarks. Our code and models are available at \href{https://zheny2751-dotcom.github.io/webvia.github.io/}{\texttt{https://webvia.github.io}}.

cs.SE↗

arXiv2025

GLM-4.5V and GLM-4.1V-Thinking: Towards Versatile Multimodal Reasoning with Scalable Reinforcement Learning

We present GLM-4.1V-Thinking, GLM-4.5V, and GLM-4.6V, a family of vision-language models (VLMs) designed to advance general-purpose multimodal understanding and reasoning. In this report, we share our key findings in the development of the reasoning-centric training framework. We first develop a capable vision foundation model with significant potential through large-scale pre-training, which arguably sets the upper bound for the final performance. We then propose Reinforcement Learning with Curriculum Sampling (RLCS) to unlock the full potential of the model, leading to comprehensive capability enhancement across a diverse range of tasks, including STEM problem solving, video understanding, content recognition, coding, grounding, GUI-based agents, and long document interpretation. In a comprehensive evaluation across 42 public benchmarks, GLM-4.5V achieves state-of-the-art performance on nearly all tasks among open-source models of similar size, and demonstrates competitive or even superior results compared to closed-source models such as Gemini-2.5-Flash on challenging tasks including Coding and GUI Agents. Meanwhile, the smaller GLM-4.1V-9B-Thinking remains highly competitive-achieving superior results to the much larger Qwen2.5-VL-72B on 29 benchmarks. We open-source both GLM-4.1V-9B-Thinking and GLM-4.5V. We further introduce the GLM-4.6V series, open-source multimodal models with native tool use and a 128K context window. A brief overview is available at https://z.ai/blog/glm-4.6v. Code, models and more information are released at https://github.com/zai-org/GLM-V.

cs.CV↗

arXiv2024

Say Your Reason: Extract Contextual Rules In Situ for Context-aware Service Recommendation

This paper introduces SayRea, an interactive system that facilitates the extraction of contextual rules for personalized context-aware service recommendations in mobile scenarios. The system monitors a user's execution of registered services on their smartphones (via accessibility service) and proactively requests a single-sentence reason from the user. By utilizing a Large Language Model (LLM), SayRea parses the reason and predicts contextual relationships between the observed service and potential contexts (such as setting the alarm clock deep in the evening). In this way, SayRea can significantly reduce the cognitive load on users in anticipating future needs and selecting contextual attributes. A 10-day field study involving 20 participants showed that SayRea accumulated an average of 62.4 rules per user and successfully recommended 45% of service usage. The participants provided positive feedback on the system's usability, interpretability, and controllability. The findings highlight SayRea's effectiveness in personalized service recommendations and its potential to enhance user experience in mobile scenarios.

cs.HC↗

arXiv2024

VisualAgentBench: Towards Large Multimodal Models as Visual Foundation Agents

Large Multimodal Models (LMMs) have ushered in a new era in artificial intelligence, merging capabilities in both language and vision to form highly capable Visual Foundation Agents. These agents are postulated to excel across a myriad of tasks, potentially approaching general artificial intelligence. However, existing benchmarks fail to sufficiently challenge or showcase the full potential of LMMs in complex, real-world environments. To address this gap, we introduce VisualAgentBench (VAB), a comprehensive and pioneering benchmark specifically designed to train and evaluate LMMs as visual foundation agents across diverse scenarios, including Embodied, Graphical User Interface, and Visual Design, with tasks formulated to probe the depth of LMMs' understanding and interaction capabilities. Through rigorous testing across nine proprietary LMM APIs and eight open models, we demonstrate the considerable yet still developing agent capabilities of these models. Additionally, VAB constructs a trajectory training set constructed through hybrid methods including Program-based Solvers, LMM Agent Bootstrapping, and Human Demonstrations, promoting substantial performance improvements in LMMs through behavior cloning. Our work not only aims to benchmark existing models but also provides a solid foundation for future development into visual foundation agents. Code, train \& test data, and part of fine-tuned open LMMs are available at \url{https://github.com/THUDM/VisualAgentBench}.

cs.AI↗

arXiv2024

A Human-Computer Collaborative Tool for Training a Single Large Language Model Agent into a Network through Few Examples

The capabilities of a single large language model (LLM) agent for solving a complex task are limited. Connecting multiple LLM agents to a network can effectively improve overall performance. However, building an LLM agent network (LAN) requires a substantial amount of time and effort. In this paper, we introduce EasyLAN, a human-computer collaborative tool that helps developers construct LANs. EasyLAN initially generates a LAN containing only one agent based on the description of the desired task. Subsequently, EasyLAN leverages a few training examples to update the LAN. For each example, EasyLAN models the gap between the output and the ground truth and identifies the causes of the errors. These errors are addressed through carefully designed strategies. Users can intervene in EasyLAN's workflow or directly modify the LAN. Eventually, the LAN evolves from a single agent to a network of LLM agents. The experimental results indicate that developers can rapidly construct LANs with good performance.

cs.HC↗

arXiv2023

AutoTask: Executing Arbitrary Voice Commands by Exploring and Learning from Mobile GUI

Voice command interfaces (VCIs) have gained increasing importance, enabling hands-free and eyes-free interaction with digital devices. However, the inherent complexity in constructing effective voice interfaces has limited the VCIs' functionalities to only a small fraction of GUI applications and tasks. This paper presents AutoTask, a VCI capable of automating any task in any mobile application without configuration or modification from developers or end users. The primary challenge for AutoTask is the lack of knowledge, as it needs to accomplish unknown tasks (e.g., user commands) within an unknown environment (e.g., GUI). To address this challenge, AutoTask employs two strategies: (1) trial and error: AutoTask explores the GUI, attempts potential operation sequences, and recovers from errors through backtracking; (2) learning from the environment: AutoTask accumulates experiences during exploration and summarizes correct knowledge from these experiences. We implemented AutoTask on Android devices and conducted an evaluation study, which proved the feasibility of AutoTask.

cs.HC↗