Hongshen Xu

Zihan Zhao, Lu Chen, Ruisheng Cao et al.

Recently, the structural reading comprehension (SRC) task on web pages has attracted increasing research interests. Although previous SRC work has leveraged extra information such as HTML tags or XPaths, the informative topology of web pages is not effectively exploited. In this work, we propose a Topological Information Enhanced model (TIE), which transforms the token-level task into a tag-level task by introducing a two-stage process (i.e. node locating and answer refining). Based on that, TIE integrates Graph Attention Network (GAT) and Pre-trained Language Model (PLM) to leverage the topological information of both logical structures and spatial structures. Experimental results demonstrate that our model outperforms strong baselines and achieves state-of-the-art performances on the web-based SRC benchmark WebSRC at the time of writing. The code of TIE will be publicly available at https://github.com/X-LANCE/TIE.

Ruisheng Cao, Fangyu Lei, Haoyuan Wu et al. · tsinghua

Data science and engineering workflows often span multiple stages, from warehousing to orchestration, using tools like BigQuery, dbt, and Airbyte. As vision language models (VLMs) advance in multimodal understanding and code generation, VLM-based agents could potentially automate these workflows by generating SQL queries, Python code, and GUI operations. This automation can improve the productivity of experts while democratizing access to large-scale data analysis. In this paper, we introduce Spider2-V, the first multimodal agent benchmark focusing on professional data science and engineering workflows, featuring 494 real-world tasks in authentic computer environments and incorporating 20 enterprise-level professional applications. These tasks, derived from real-world use cases, evaluate the ability of a multimodal agent to perform data-related tasks by writing code and managing the GUI in enterprise data software systems. To balance realistic simulation with evaluation simplicity, we devote significant effort to developing automatic configurations for task setup and carefully crafting evaluation metrics for each task. Furthermore, we supplement multimodal agents with comprehensive documents of these enterprise data software systems. Our empirical evaluation reveals that existing state-of-the-art LLM/VLM-based agents do not reliably automate full data workflows (14.0% success). Even with step-by-step guidance, these agents still underperform in tasks that require fine-grained, knowledge-intensive GUI actions (16.2%) and involve remote cloud-hosted workspaces (10.6%). We hope that Spider2-V paves the way for autonomous multimodal agents to transform the automation of data science and engineering workflow. Our code and data are available at https://spider2-v.github.io.

Danyang Zhang, Lu Chen, Situo Zhang et al.

Inspired by the insights in cognitive science with respect to human memory and reasoning mechanism, a novel evolvable LLM-based (Large Language Model) agent framework is proposed as REMEMBERER. By equipping the LLM with a long-term experience memory, REMEMBERER is capable of exploiting the experiences from the past episodes even for different task goals, which excels an LLM-based agent with fixed exemplars or equipped with a transient working memory. We further introduce Reinforcement Learning with Experience Memory (RLEM) to update the memory. Thus, the whole system can learn from the experiences of both success and failure, and evolve its capability without fine-tuning the parameters of the LLM. In this way, the proposed REMEMBERER constitutes a semi-parametric RL agent. Extensive experiments are conducted on two RL task sets to evaluate the proposed framework. The average results with different initialization and training sets exceed the prior SOTA by 4% and 2% for the success rate on two task sets and demonstrate the superiority and robustness of REMEMBERER.

Xiaomi LLM-Core Team, Dong Zhang, Gang Wang et al.

Existing audio language models typically rely on task-specific fine-tuning to accomplish particular audio tasks. In contrast, humans are able to generalize to new audio tasks with only a few examples or simple instructions. GPT-3 has shown that scaling next-token prediction pretraining enables strong generalization capabilities in text, and we believe this paradigm is equally applicable to the audio domain. By scaling MiMo-Audio's pretraining data to over one hundred million of hours, we observe the emergence of few-shot learning capabilities across a diverse set of audio tasks. We develop a systematic evaluation of these capabilities and find that MiMo-Audio-7B-Base achieves SOTA performance on both speech intelligence and audio understanding benchmarks among open-source models. Beyond standard metrics, MiMo-Audio-7B-Base generalizes to tasks absent from its training data, such as voice conversion, style transfer, and speech editing. MiMo-Audio-7B-Base also demonstrates powerful speech continuation capabilities, capable of generating highly realistic talk shows, recitations, livestreaming and debates. At the post-training stage, we curate a diverse instruction-tuning corpus and introduce thinking mechanisms into both audio understanding and generation. MiMo-Audio-7B-Instruct achieves open-source SOTA on audio understanding benchmarks (MMSU, MMAU, MMAR, MMAU-Pro), spoken dialogue benchmarks (Big Bench Audio, MultiChallenge Audio) and instruct-TTS evaluations, approaching or surpassing closed-source models. Model checkpoints and full evaluation suite are available at https://github.com/XiaomiMiMo/MiMo-Audio.

Hanchong Zhang, Ruisheng Cao, Lu Chen et al.

Recently Large Language Models (LLMs) have been proven to have strong abilities in various domains and tasks. We study the problem of prompt designing in the text-to-SQL task and attempt to improve the LLMs' reasoning ability when generating SQL queries. Besides the trivial few-shot in-context learning setting, we design our chain-of-thought (CoT) prompt with a similar method to schema linking. We provide a method named ACT-SQL to automatically generate auto-CoT exemplars and thus the whole process doesn't need manual labeling. Our approach is cost-saving since we only use the LLMs' API call once when generating one SQL query. Furthermore, we extend our in-context learning method to the multi-turn text-to-SQL task. The experiment results show that the LLMs' performance can benefit from our ACT-SQL approach. Our approach achieves SOTA performance on the Spider dev set among existing in-context learning approaches.

Ruisheng Cao, Hanchong Zhang, Hongshen Xu et al.

Text-to-SQL aims to generate an executable SQL program given the user utterance and the corresponding database schema. To ensure the well-formedness of output SQLs, one prominent approach adopts a grammar-based recurrent decoder to produce the equivalent SQL abstract syntax tree (AST). However, previous methods mainly utilize an RNN-series decoder, which 1) is time-consuming and inefficient and 2) introduces very few structure priors. In this work, we propose an AST structure-aware Transformer decoder (ASTormer) to replace traditional RNN cells. The structural knowledge, such as node types and positions in the tree, is seamlessly incorporated into the decoder via both absolute and relative position embeddings. Besides, the proposed framework is compatible with different traversing orders even considering adaptive node selection. Extensive experiments on five text-to-SQL benchmarks demonstrate the effectiveness and efficiency of our structured decoder compared to competitive baselines.

Jieyu Li, Lu Chen, Ruisheng Cao et al.

Exploring the generalization of a text-to-SQL parser is essential for a system to automatically adapt the real-world databases. Previous works provided investigations focusing on lexical diversity, including the influence of the synonym and perturbations in both natural language questions and databases. However, research on the structure variety of database schema~(DS) is deficient. Specifically, confronted with the same input question, the target SQL is probably represented in different ways when the DS comes to a different structure. In this work, we provide in-deep discussions about the structural generalization of text-to-SQL tasks. We observe that current datasets are too templated to study structural generalization. To collect eligible test data, we propose a framework to generate novel text-to-SQL data via automatic and synchronous (DS, SQL) pair altering. In the experiments, significant performance reduction when evaluating well-trained text-to-SQL models on the synthetic samples demonstrates the limitation of current research regarding structural generalization. According to comprehensive analysis, we suggest the practical reason is the overfitting of (NL, SQL) patterns.

LLM-Core Xiaomi, Bingquan Xia, Bowen Shen et al. · pku

We present MiMo-7B, a large language model born for reasoning tasks, with optimization across both pre-training and post-training stages. During pre-training, we enhance the data preprocessing pipeline and employ a three-stage data mixing strategy to strengthen the base model's reasoning potential. MiMo-7B-Base is pre-trained on 25 trillion tokens, with additional Multi-Token Prediction objective for enhanced performance and accelerated inference speed. During post-training, we curate a dataset of 130K verifiable mathematics and programming problems for reinforcement learning, integrating a test-difficulty-driven code-reward scheme to alleviate sparse-reward issues and employing strategic data resampling to stabilize training. Extensive evaluations show that MiMo-7B-Base possesses exceptional reasoning potential, outperforming even much larger 32B models. The final RL-tuned model, MiMo-7B-RL, achieves superior performance on mathematics, code and general reasoning tasks, surpassing the performance of OpenAI o1-mini. The model checkpoints are available at https://github.com/xiaomimimo/MiMo.

Xiaomi LLM-Core Team, Zihao Yue, Zhenru Lin et al. · pku

We open-source MiMo-VL-7B-SFT and MiMo-VL-7B-RL, two powerful vision-language models delivering state-of-the-art performance in both general visual understanding and multimodal reasoning. MiMo-VL-7B-RL outperforms Qwen2.5-VL-7B on 35 out of 40 evaluated tasks, and scores 59.4 on OlympiadBench, surpassing models with up to 78B parameters. For GUI grounding applications, it sets a new standard with 56.1 on OSWorld-G, even outperforming specialized models such as UI-TARS. Our training combines four-stage pre-training (2.4 trillion tokens) with Mixed On-policy Reinforcement Learning (MORL) integrating diverse reward signals. We identify the importance of incorporating high-quality reasoning data with long Chain-of-Thought into pre-training stages, and the benefits of mixed RL despite challenges in simultaneous multi-domain optimization. We also contribute a comprehensive evaluation suite covering 50+ tasks to promote reproducibility and advance the field. The model checkpoints and full evaluation suite are available at https://github.com/XiaomiMiMo/MiMo-VL.

Hongshen Xu, Lu Chen, Zihan Zhao et al.

The growing prevalence of visually rich documents, such as webpages and scanned/digital-born documents (images, PDFs, etc.), has led to increased interest in automatic document understanding and information extraction across academia and industry. Although various document modalities, including image, text, layout, and structure, facilitate human information retrieval, the interconnected nature of these modalities presents challenges for neural networks. In this paper, we introduce WebLM, a multimodal pre-training network designed to address the limitations of solely modeling text and structure modalities of HTML in webpages. Instead of processing document images as unified natural images, WebLM integrates the hierarchical structure of document images to enhance the understanding of markup-language-based documents. Additionally, we propose several pre-training tasks to model the interaction among text, structure, and image modalities effectively. Empirical results demonstrate that the pre-trained WebLM significantly surpasses previous state-of-the-art pre-trained models across several webpage understanding tasks. The pre-trained models and code are available at https://github.com/X-LANCE/weblm.

Da Ma, Lu Chen, Pengyu Wang et al.

Large language models (LLMs) have demonstrated proficiency across various natural language processing (NLP) tasks but often require additional training, such as continual pre-training and supervised fine-tuning. However, the costs associated with this, primarily due to their large parameter count, remain high. This paper proposes leveraging \emph{sparsity} in pre-trained LLMs to expedite this training process. By observing sparsity in activated neurons during forward iterations, we identify the potential for computational speed-ups by excluding inactive neurons. We address associated challenges by extending existing neuron importance evaluation metrics and introducing a ladder omission rate scheduler. Our experiments on Llama-2 demonstrate that Sparsity-Accelerated Training (SAT) achieves comparable or superior performance to standard training while significantly accelerating the process. Specifically, SAT achieves a $45\%$ throughput improvement in continual pre-training and saves $38\%$ training time in supervised fine-tuning in practice. It offers a simple, hardware-agnostic, and easily deployable framework for additional LLM training. Our code is available at https://github.com/OpenDFM/SAT.

Zihan Zhao, Da Ma, Lu Chen et al.

Artificial intelligence (AI) has played an increasingly important role in chemical research. However, most models currently used in chemistry are specialist models that require training and tuning for specific tasks. A more generic and efficient solution would be an AI model that could address many tasks and support free-form dialogue in the broad field of chemistry. In its utmost form, such a generalist AI chemist could be referred to as Chemical General Intelligence. Large language models (LLMs) have recently logged tremendous success in the general domain of natural language processing, showing emerging task generalization and free-form dialogue capabilities. However, domain knowledge of chemistry is largely missing when training general-domain LLMs. The lack of such knowledge greatly hinders the performance of generalist LLMs in the field of chemistry. To this end, we develop ChemDFM, a pioneering LLM for chemistry trained on 34B tokens from chemical literature and textbooks, and fine-tuned using 2.7M instructions. As a result, it can understand and reason with chemical knowledge in free-form dialogue. Quantitative evaluations show that ChemDFM significantly surpasses most representative open-source LLMs. It outperforms GPT-4 on a great portion of chemical tasks, despite the substantial size difference. We have open-sourced the inference codes, evaluation datasets, and model weights of ChemDFM on Huggingface (https://huggingface.co/OpenDFM/ChemDFM-v1.0-13B).

Hongshen Xu, Zichen Zhu, Situo Zhang et al.

Large Language Models (LLMs) often generate erroneous outputs, known as hallucinations, due to their limitations in discerning questions beyond their knowledge scope. While addressing hallucination has been a focal point in research, previous efforts primarily concentrate on enhancing correctness without giving due consideration to the significance of rejection mechanisms. In this paper, we conduct a comprehensive examination of the role of rejection, introducing the notion of model reliability along with corresponding metrics. These metrics measure the model's ability to provide accurate responses while adeptly rejecting questions exceeding its knowledge boundaries, thereby minimizing hallucinations. To improve the inherent reliability of LLMs, we present a novel alignment framework called Reinforcement Learning from Knowledge Feedback (RLKF). RLKF leverages knowledge feedback to dynamically determine the model's knowledge boundary and trains a reliable reward model to encourage the refusal of out-of-knowledge questions. Experimental results on mathematical questions affirm the substantial efficacy of RLKF in significantly enhancing LLM reliability.

Hongshen Xu, Zichen Zhu, Lei Pan et al.

Large Language Models (LLMs) have expanded their capabilities beyond language generation to interact with external tools, enabling automation and real-world applications. However, tool hallucinations, where models either select inappropriate tools or misuse them, pose significant challenges, leading to erroneous task execution, increased computational costs, and reduced system reliability. To systematically address this issue, we define and categorize tool hallucinations into two main types, tool selection hallucination and tool usage hallucination. To evaluate and mitigate these issues, we introduce RelyToolBench, which integrates specialized test cases and novel metrics to assess hallucination-aware task success and efficiency. Finally, we propose Relign, a reliability alignment framework that expands the tool-use action space to include indecisive actions, allowing LLMs to defer tool use, seek clarification, or adjust tool selection dynamically. Through extensive experiments, we demonstrate that Relign significantly reduces tool hallucinations, improves task reliability, and enhances the efficiency of LLM tool interactions.

Hanchong Zhang, Ruisheng Cao, Hongshen Xu et al.

Recently, Large Language Models (LLMs) have been demonstrated to possess impressive capabilities in a variety of domains and tasks. We investigate the issue of prompt design in the multi-turn text-to-SQL task and attempt to enhance the LLMs' reasoning capacity when generating SQL queries. In the conversational context, the current SQL query can be modified from the preceding SQL query with only a few operations due to the context dependency. We introduce our method called CoE-SQL which can prompt LLMs to generate the SQL query based on the previously generated SQL query with an edition chain. We also conduct extensive ablation studies to determine the optimal configuration of our approach. Our approach outperforms different in-context learning baselines stably and achieves state-of-the-art performances on two benchmarks SParC and CoSQL using LLMs, which is also competitive to the SOTA fine-tuned models.

Hongchuan Zeng, Hongshen Xu, Lu Chen et al.

Large Language Models (LLMs) have ushered in a new era in Natural Language Processing, but their massive size demands effective compression techniques for practicality. Although numerous model compression techniques have been investigated, they typically rely on a calibration set that overlooks the multilingual context and results in significant accuracy degradation for low-resource languages. This paper introduces Multilingual Brain Surgeon (MBS), a novel calibration data sampling method for multilingual LLMs compression. MBS overcomes the English-centric limitations of existing methods by sampling calibration data from various languages proportionally to the language distribution of the model training datasets. Our experiments, conducted on the BLOOM multilingual LLM, demonstrate that MBS improves the performance of existing English-centric compression methods, especially for low-resource languages. We also uncover the dynamics of language interaction during compression, revealing that the larger the proportion of a language in the training set and the more similar the language is to the calibration language, the better performance the language retains after compression. In conclusion, MBS presents an innovative approach to compressing multilingual LLMs, addressing the performance disparities and improving the language inclusivity of existing compression techniques.

Zichen Zhu, Hao Tang, Yansi Li et al.

Existing Multimodal Large Language Model (MLLM)-based agents face significant challenges in handling complex GUI (Graphical User Interface) interactions on devices. These challenges arise from the dynamic and structured nature of GUI environments, which integrate text, images, and spatial relationships, as well as the variability in action spaces across different pages and tasks. To address these limitations, we propose MobA, a novel MLLM-based mobile assistant system. MobA introduces an adaptive planning module that incorporates a reflection mechanism for error recovery and dynamically adjusts plans to align with the real environment contexts and action module's execution capacity. Additionally, a multifaceted memory module provides comprehensive memory support to enhance adaptability and efficiency. We also present MobBench, a dataset designed for complex mobile interactions. Experimental results on MobBench and AndroidArena demonstrate MobA's ability to handle dynamic GUI environments and perform complex mobile tasks.

Hang Zheng, Hongshen Xu, Yuncong Liu et al.

Large language models (LLMs) are prone to hallucination stemming from misaligned self-awareness, particularly when processing queries exceeding their knowledge boundaries. While existing mitigation strategies employ uncertainty estimation or query rejection mechanisms, they suffer from computational efficiency and sacrificed helpfulness. To address these issues, we propose the Explicit Knowledge Boundary Modeling (EKBM) framework, integrating fast and slow reasoning systems to harmonize reliability and usability. The framework first employs a fast-thinking model to generate confidence-labeled responses, enabling immediate utilization of high-confidence outputs, whereas uncertain predictions trigger a slow refinement model for accuracy improvement. To align model behavior with our proposed object, we propose a hybrid training pipeline, enhancing self-awareness without degrading task performance. Evaluations on dialogue state tracking tasks demonstrate that EKBM achieves superior model reliability over uncertainty-based baselines. Further analysis reveals that refinement substantially boosts accuracy while maintaining low computational overhead. The framework establishes a scalable paradigm for deploying reliable LLMs in error-sensitive applications, effectively balancing accuracy and practical utility.

Da Ma, Lu Chen, Situo Zhang et al.

The rapid expansion of context window sizes in Large Language Models~(LLMs) has enabled them to tackle increasingly complex tasks involving lengthy documents. However, this progress comes at the cost of a substantial increase in memory usage during inference, primarily due to the linear growth of the key-value~(KV) cache. Existing KV cache compression methods often discard less relevant tokens, which can lead to significant performance degradation when critical information is lost. In this paper, we propose \textsc{PoD}~(Proximal tokens over Distant tokens), a novel KV cache compression framework that allocates memory according to token importance, retaining less important tokens in a more compact, shared form rather than discarding them entirely. Our approach is motivated by two key observations: (1) proximal tokens -- those at the beginning and end of the context -- are significantly more important for next-token prediction, and (2) attention scores for distant tokens are highly redundant across consecutive layers. Leveraging these insights, \textsc{PoD} preserves the full KV cache for proximal tokens, while for distant tokens, it shares key states across layers. Since attention scores are determined by both queries and keys, sharing key states enables multiple layers to reuse a single set of keys for distant tokens, substantially reducing KV cache memory without discarding essential context. We further introduce a lightweight post-training adaptation to enable the model to adjust to this new attention-sharing structure. Extensive experiments on both synthetic~(Needle in a Haystack) and real-world long-context benchmarks demonstrate that \textsc{PoD} reduces KV cache memory usage by up to 35\% without compromising performance. Our method is orthogonal to existing token-selection-based techniques and can be combined with them for further KV cache compression.

Hongshen Xu, Zihan Wang, Zichen Zhu et al.

Recent advancements in tool learning have enabled large language models (LLMs) to integrate external tools, enhancing their task performance by expanding their knowledge boundaries. However, relying on tools often introduces tradeoffs between performance, speed, and cost, with LLMs sometimes exhibiting overreliance and overconfidence in tool usage. This paper addresses the challenge of aligning LLMs with their knowledge boundaries to make more intelligent decisions about tool invocation. We propose a multi objective alignment framework that combines probabilistic knowledge boundary estimation with dynamic decision making, allowing LLMs to better assess when to invoke tools based on their confidence. Our framework includes two methods for knowledge boundary estimation, consistency based and absolute estimation, and two training strategies for integrating these estimates into the model decision making process. Experimental results on various tool invocation scenarios demonstrate the effectiveness of our framework, showing significant improvements in tool efficiency by reducing unnecessary tool usage.

Hongshen Xu, Ruisheng Cao, Su Zhu et al.

Previous work on spoken language understanding (SLU) mainly focuses on single-intent settings, where each input utterance merely contains one user intent. This configuration significantly limits the surface form of user utterances and the capacity of output semantics. In this work, we first propose a Multi-Intent dataset which is collected from a realistic in-Vehicle dialogue System, called MIVS. The target semantic frame is organized in a 3-layer hierarchical structure to tackle the alignment and assignment problems in multi-intent cases. Accordingly, we devise a BiRGAT model to encode the hierarchy of ontology items, the backbone of which is a dual relational graph attention network. Coupled with the 3-way pointer-generator decoder, our method outperforms traditional sequence labeling and classification-based schemes by a large margin.

Hang Zheng, Hongshen Xu, Yongkai Lin et al.

The proliferation of Large Language Models (LLMs) has created a diverse ecosystem of models with highly varying performance and costs, necessitating effective query routing to balance performance and expense. Current routing systems often rely on a centralized external router trained on a fixed set of LLMs, making them inflexible and prone to poor performance since the small router can not fully understand the knowledge boundaries of different LLMs. We introduce DiSRouter (Distributed Self-Router), a novel paradigm that shifts from centralized control to distributed routing. In DiSRouter, a query traverses a network of LLM agents, each independently deciding whether to answer or route to other agents based on its own self-awareness, its ability to judge its competence. This distributed design offers superior flexibility, scalability, and generalizability. To enable this, we propose a two-stage Self-Awareness Training pipeline that enhances each LLM's self-awareness. Extensive experiments demonstrate that DiSRouter significantly outperforms existing routing methods in utility across various scenarios, effectively distinguishes between easy and hard queries, and shows strong generalization to out-of-domain tasks. Our work validates that leveraging an LLM's intrinsic self-awareness is more effective than external assessment, paving the way for more modular and efficient multi-agent systems.

Xinzhe Xu, Liang Zhao, Hongshen Xu et al.

Large Language Models (LLMs) are increasingly tasked with analyzing legal texts and citing relevant statutes, yet their reliability is often compromised by general pre-training that ingests legal texts without specialized focus, obscuring the true depth of their legal knowledge. This paper introduces CLaw, a novel benchmark specifically engineered to meticulously evaluate LLMs on Chinese legal knowledge and its application in reasoning. CLaw comprises two key components: (1) a comprehensive, fine-grained corpus of all 306 Chinese national statutes, segmented to the subparagraph level and incorporating precise historical revision timesteps for rigorous recall evaluation (64,849 entries), and (2) a challenging set of 254 case-based reasoning instances derived from China Supreme Court curated materials to assess the practical application of legal knowledge. Our empirical evaluation reveals that most contemporary LLMs significantly struggle to faithfully reproduce legal provisions. As accurate retrieval and citation of legal provisions form the basis of legal reasoning, this deficiency critically undermines the reliability of their responses. We contend that achieving trustworthy legal reasoning in LLMs requires a robust synergy of accurate knowledge retrieval--potentially enhanced through supervised fine-tuning (SFT) or retrieval-augmented generation (RAG)--and strong general reasoning capabilities. This work provides an essential benchmark and critical insights for advancing domain-specific LLM reasoning, particularly within the complex legal sphere.

Hongshen Xu, Zixv yang, Zichen Zhu et al.

Large Language Models often generate factually incorrect but plausible outputs, known as hallucinations. We identify a more insidious phenomenon, LLM delusion, defined as high belief hallucinations, incorrect outputs with abnormally high confidence, making them harder to detect and mitigate. Unlike ordinary hallucinations, delusions persist with low uncertainty, posing significant challenges to model reliability. Through empirical analysis across different model families and sizes on several Question Answering tasks, we show that delusions are prevalent and distinct from hallucinations. LLMs exhibit lower honesty with delusions, which are harder to override via finetuning or self reflection. We link delusion formation with training dynamics and dataset noise and explore mitigation strategies such as retrieval augmented generation and multi agent debating to mitigate delusions. By systematically investigating the nature, prevalence, and mitigation of LLM delusions, our study provides insights into the underlying causes of this phenomenon and outlines future directions for improving model reliability.

Danyang Zhang, Zhennan Shen, Rui Xie et al.

The Graphical User Interface (GUI) is pivotal for human interaction with the digital world, enabling efficient device control and the completion of complex tasks. Recent progress in Large Language Models (LLMs) and Vision Language Models (VLMs) offers the chance to create advanced GUI agents. To ensure their effectiveness, there's a pressing need for qualified benchmarks that provide trustworthy and reproducible evaluations -- a challenge current benchmarks often fail to address. To tackle this issue, we introduce Mobile-Env, a comprehensive toolkit tailored for creating GUI benchmarks in the Android mobile environment. Mobile-Env offers an isolated and controllable setting for reliable evaluations, and accommodates intermediate instructions and rewards to reflect real-world usage more naturally. Utilizing Mobile-Env, we collect an open-world task set across various real-world apps and a fixed world set, WikiHow, which captures a significant amount of dynamic online contents for fully controllable and reproducible evaluation. We conduct comprehensive evaluations of LLM agents using these benchmarks. Our findings reveal that even advanced models (e.g., GPT-4V and LLaMA-3) struggle with tasks that are relatively simple for humans. This highlights a crucial gap in current models and underscores the importance of developing more capable foundation models and more effective GUI agent frameworks.