Pith. sign in

REVIEW 5 major objections 6 minor 6 cited by

MobileGUI-RL: Advancing Mobile GUI Agent through Reinforcement Learning in Online Environment

T0 review · 5 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read MobileGUI-RL claims that online reinforcement learning, with self-explored task curricula and trajectory-level rewards, lifts mobile GUI agents by up to 13.3 points over their base checkpoints.

desk verdict Promising online RL pipeline for mobile GUI agents, but the training AVD is named AndroidWorldAvd and the paper never shows the 436 training tasks are disjoint from AndroidWorld eval tasks — that missing guarantee has to be resolved before the headline numbers mean anything. read the letter →

arxiv 2507.05720 v1 pith:W3SAI6NO submitted 2025-07-08 cs.LG cs.CL

Yucheng Shi , Wenhao Yu , Zaitang Li , Yonglin Wang , Hongming Zhang , Ninghao Liu , Haitao Mi , Dong Yu This is my paper
classification cs.LGcs.CL
keywords mobileGUIagentsonlinereinforcementlearningGRPOcurriculumvision-languagemodelsAndroidWorldtaskfilteringtrajectory-levelreward
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper tries to establish that vision-based mobile GUI agents can be trained effectively through reinforcement learning in an online emulator environment, rather than by fine-tuning on pre-collected offline trajectories. The authors argue that offline training overfits to specific interface templates and fails on unseen screens, while online interaction lets a policy adapt to live GUI dynamics. Their framework, MobileGUI-RL, generates tasks by reverse-engineering natural-language instructions from exploratory trajectories, filters those tasks through a text-based world model to keep only solvable ones, and then optimizes with a GRPO-style algorithm modified for trajectory-level rewards. The central empirical claim is that this yields large, consistent gains: MobileGUI-32B reaches 44.8% success on AndroidWorld, outperforming GPT-4o (34.5%) and the larger Qwen2.5-VL-72B (35.0%), while improving 13.3 points over its own 32B base model. If true, this points to a scalable path for training capable GUI agents without hand-annotated action trajectories.

What carries the argument

The load-bearing machinery is MobGRPO, an adaptation of group relative policy optimization (GRPO) for long-horizon GUI navigation. Instead of per-step rewards, it assigns a single normalized advantage to every action in a completed trajectory, computed from the trajectory's composite reward relative to other rollouts of the same task. The composite reward combines a binary success signal from a vision-language oracle, an exponentially decaying efficiency factor that rewards shorter successful trajectories, and a linear penalty that discourages premature termination; degenerate batches where all rollouts fail are filtered out following the DAPO recipe. A second key mechanism is the task-generation and filtering pipeline: a GPT-4o-based reverse task synthesizer converts random exploration trajectories into instructions, and a text-based world model W simulates whether a proxy agent can solve each candidate task, admitting only feasible tasks into the curriculum and ordering them by estimated complexity.

What would settle it

Retrain MobileGUI-32B on the same 436-task curriculum and rollouts but replace the vision-language oracle's success judgments with human annotations; if AndroidWorld success no longer reaches roughly 44.8%, the reported gain depends on evaluator bias rather than genuine task completion.

Watch

Extended reading notes

Core claim

The paper's central discovery is that online reinforcement learning can be made to work for mobile GUI agents if three design choices are combined: self-exploration for task discovery, a lightweight text-based world model for task filtering and curriculum construction, and a trajectory-level advantage with a composite reward that rewards success, penalizes premature termination, and distinguishes efficient from inefficient successful trajectories. The authors adapt GRPO to GUI navigation, calling their variant MobGRPO, and show through ablations that each component contributes: removing task filtering drops 32B AndroidWorld success from 44.8% to 41.0%, removing curriculum learning drops it to 34.0%, and removing the decaying efficiency reward drops it to 35.5%. The resulting agents improve over their base models on all three evaluated benchmarks, with the largest gains on the more diverse Android-in-the-Wild tasks.

Load-bearing premise

The success signal and task-feasibility filter are trusted: a vision-language oracle decides whether a trajectory completed a task, and a text-based world model decides which tasks are learnable, with no human validation reported.

Editorial extensions

If this is right

  • If correct, a 32B open-weights model can surpass both its 72B sibling and a leading closed-source model on AndroidWorld without task-specific reward engineering or human action annotations.
  • The same recipe transfers to 7B models, where AndroidWorld success rises from 22.0% to 30.0% and AITW-Gen from 49.0% to 65.3%.
  • Removing the task filter, the curriculum, or the decaying reward each causes clear degradation, so all three components are load-bearing for the reported gains.
  • Training curves show steady online improvement, suggesting the framework could support continual adaptation as new tasks or UI layouts appear.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A testable extension is to replace the proprietary LLM used for reverse task synthesis with the agent's own policy, which would remove the last external dependency and test whether the curriculum can be fully self-generated.
  • The text-based world-model filter may generalize to web or desktop GUI environments, but the paper only demonstrates it on Android emulators, so that transfer is an open question.
  • Because the paper does not report human validation of the oracle's success judgments, a direct audit of oracle accuracy on held-out trajectories would clarify how much of the measured gain is genuine task completion rather than evaluator bias.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 6 minor

Summary. This paper introduces MobileGUI-RL, a framework for training mobile GUI agents via online reinforcement learning. It combines self-exploration to synthesize tasks from random-walk trajectories with GPT-4o reverse instruction generation, a text-based world model that filters tasks for feasibility, and an adapted GRPO objective (MobGRPO) with trajectory-level advantages and a composite reward balancing success, efficiency, and premature-termination penalty. The authors train Qwen2.5-VL-7B and 32B on 436 generated tasks and evaluate on AndroidWorld, AITW-Gen, and AITW-Web. They report substantial gains over base models, including 44.8% vs 35.0% for Qwen2.5-VL-72B on AndroidWorld, and claim state-of-the-art results among compared systems.

Significance. The claimed results are significant if substantiated: they would demonstrate that online RL with self-generated curricula and trajectory-level rewards can improve vision-language GUI agents beyond offline SFT and prior RL methods, and that a 32B model can surpass larger closed/open models on AndroidWorld. The paper is strong in laying out a complete pipeline with public benchmarks and ablations of filtering, curriculum, and reward shaping. However, the significance is conditional on resolving evaluation comparability and supervision-validity concerns; as written, the empirical evidence does not yet support the strongest claims.

major comments (5)
  1. [Section 3.3.1 / Appendix A.1.1 / Table 2] The self-exploration training environment is based on AndroidWorldAvd, and the generated curriculum is built from random walks in that same emulator/app ecosystem; the paper never states that the 436 curated training tasks are disjoint from, or distributionally different from, the AndroidWorld evaluation tasks. If the reverse-engineered instructions (e.g., calendar, camera, map tasks) overlap with AndroidWorld's test tasks, the reported AndroidWorld SR of 44.8% and the 13.3-point gain over Qwen2.5-VL-32B are confounded by training on the evaluation distribution. Please provide an explicit task-overlap analysis, or change the training environment/apps to be disjoint from evaluation, or otherwise justify that the training distribution does not cover the evaluation tasks.
  2. [Appendix C / Table 2] For AITW-Gen the evaluation set was manually reduced to 300 tasks and for AITW-Web to 150 tasks, but the paper does not state whether the baseline numbers in Table 2 (e.g., OS-Atlas 15.7/17.3, Aguvis 23.0/4.7, Qwen2.5-VL variants) were rerun on these exact curated subsets. If the baselines are taken from previously published full-set evaluations, the comparison is not apples-to-apples and the reported gains on these benchmarks may be inflated. Please rerun all baselines on the identical curated subsets or clearly state which numbers were obtained on which task sets.
  3. [Section 3.2 / Eq. (4) / Table 3] Training rewards are based on a single VLM oracle O (Qwen2.5-VL-72B) judging success from the final screenshots, with no reported human validation or agreement rate. Since Eq. (4) defines the RL reward on this binary judgment, any systematic bias in the oracle (e.g., accepting visually plausible but semantically wrong outcomes) directly undermines the learned policy and the ablation conclusions. Please report oracle accuracy against human labels on a sampled set of trajectories, and if possible compare oracle judgment with the official AndroidWorld evaluator.
  4. [Section 4.2 / Table 2] All results appear to come from a single training run; no multiple seeds, standard deviations, or significance tests are reported. Several comparisons rely on small margins (e.g., MobileGUI-7B 30.0% vs UI-TARS-7B 33.0% on AndroidWorld; MobileGUI-32B 30.7% vs Qwen2.5-VL-72B 31.3% on AITW-Web). Without variance estimates, the claim of 'consistent and significant gains' is not fully supported. Please add at least 3 seeds for the main results and the ablations in Table 3, or temper the strength of the claims.
  5. [Section 3.3.2] The text-based world model W is used to decide whether a generated task is admissible, but its fidelity to the real emulator is never validated. If W predicts transitions optimistically, the filter will admit infeasible tasks and the subsequent real-environment rollouts will be misaligned with the simulated feasibility signal. Please validate W on a sample by comparing its predicted next-state descriptions with actual emulator states after the same action, and report the correlation between W-based feasibility and real rollout success.
minor comments (6)
  1. [Introduction] Section 1 says 'Experiments on four mobile agent benchmarks', but the abstract and Table 2 describe three benchmarks; please correct the inconsistency.
  2. [Section 4.1.1 / Appendix A.2.3] Section 4.1.1 states a maximum episode length of 25 steps, while Appendix A.2.3 lists 'Maximum Steps per Episode: 15'; please reconcile these numbers.
  3. [Section 3.2] The text refers to 'appendix X' for input-construction details, but no such appendix section is present; please add the section or remove the reference.
  4. [Table 1 / Section 3.2] Table 1 lists an 'answer' action, but the action-space description in Section 3.2 enumerates only four categories (gestures, text input, system navigation, control); please align the two descriptions.
  5. [Table 2 / References] In Table 2, Aguvis-7B and Aguvis-72B are attributed to Huang et al. 2024, but that reference is a survey of LLM planning; the Aguvis paper is Xu et al. 2025. Please correct the citation.
  6. [Eq. (1)] Equation (1) uses the notation ∑G_{t=1}|oi,s| and nested sums over s and t in a way that is hard to parse; please define all indices or simplify the expression.

Circularity Check

0 steps flagged · score 2.0 of 10

No circular derivation; training/evaluation metrics are separated, and the headline gain is not defined by its inputs. Minor non-load-bearing self-citations, plus an unproven training/eval overlap risk, keep the score at 2.

full rationale

Training and evaluation are cleanly separated. Self-exploration (Sec. 3.3.1) reverse-synthesizes task instructions from emulator trajectories, the text-based world model (Sec. 3.3.2) filters tasks via a textual proxy agent, and the reward signal is produced by a VLM oracle (Sec. 3.2). None of these components is reused as the evaluation metric: AndroidWorld uses the official evaluator (Appendix C), and AITW-Gen/Web use DigiRL-derived execution scripts, not the training oracle. The MobGRPO loss (Eq. 1) and composite reward (Eq. 4) are GRPO with a trajectory-level advantage and standard reward shaping; no fitted parameter is renamed as a prediction. The paper does cite prior same-group work (WebEvolver, WebCoT, rollback mechanisms) for the MDP convention and related ideas, but those citations are background and not load-bearing premises of the headline result. The remaining concerns — the training AVD is named AndroidWorldAvd (Appendix A.1.1) while AndroidWorld is also an evaluation benchmark, and the AITW subsets are manually curated (Appendix C) — are omitted-guarantee/correctness risks, not demonstrated circular reductions, so under the hard rules they do not raise the circularity score beyond the minor self-citation level.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The central claim depends mainly on unvalidated reward and feasibility signals, plus several unreported reward hyperparameters. No new physical or conceptual entities such as particles, forces, or dimensions are introduced; the text-based world model and oracle are software components, not postulated entities.

free parameters (5)
  • lambda (efficiency decay rate) = not reported
    Controls reward decay for longer successful trajectories in Eq. (2); no value or sensitivity analysis is given.
  • alpha_min, alpha_max (efficiency clip bounds) = not reported
    Clip the efficiency factor in Eq. (2); without values the reward magnitude is unspecified.
  • beta_max (premature termination penalty weight) = not reported
    Scales the failure penalty in Eq. (3); no value or sensitivity analysis is given.
  • rbase (success base reward) = not stated (likely 1)
    Multiplies the efficiency factor on success in Eq. (4); not listed in the appendix hyperparameters.
  • Tmax (maximum episode length) = 15 in Appendix A.2.3, 25 in Section 4.1.1
    Appears in the failure penalty and filtering step limit; the two reported values are inconsistent.
assumptions (5)
  • domain assumption GRPO with a single trajectory-level advantage assigned to all tokens is a valid optimization signal for long-horizon GUI tasks.
    Eq. (1) extends GRPO without proof or comparison to step-level credit assignment; Section 6 itself notes that trajectory-level rewards can produce ambiguous learning signals.
  • domain assumption The text-based world model W accurately simulates the effect of actions on GUI state.
    Section 3.3.2 admits tasks only if a simulated proxy run succeeds; no accuracy measurement against real emulator rollouts is provided.
  • domain assumption The VLM oracle O produces correct binary success judgments.
    Section 3.2 uses Qwen2.5-VL-72B as a unified evaluator with no human agreement study or error analysis.
  • domain assumption Reverse-engineered tasks from GPT-4o correspond to executable goals in the emulator.
    Section 3.3.1 relies on GPT-4o to infer instructions from random walks; the text-based filter is the only feasibility check.
  • domain assumption Android emulator behavior is representative of real mobile device GUI dynamics.
    Experiments use AndroidWorld AVDs; no real-device validation is reported.

how reviews work

0 comments
Cite this review

Pith. "Pith review of MobileGUI-RL: Advancing Mobile GUI Agent through Reinforcement Learning in Online Environment." pith.science (2026). https://pith.science/paper/W3SAI6NO

@misc{pith2026250705720,
  author       = {Pith},
  title        = {Pith review of: MobileGUI-RL: Advancing Mobile GUI Agent through Reinforcement Learning in Online Environment},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/W3SAI6NO}},
  note         = {Machine review of arXiv:2507.05720}
}
read the original abstract

Recently, there has been a surge of vision-based GUI agents designed to automate everyday mobile and web tasks. These agents interpret raw GUI screenshots and autonomously decide where to click, scroll, or type, which bypasses handcrafted rules and app-specific APIs. However, most existing methods trained GUI agent in the offline environment using pre-collected trajectories. This approach limits scalability, causes overfitting to specific UI templates, and leads to brittle policies when faced with unseen environment. We present MobileGUI-RL, a scalable framework that trains GUI agent in online environment. MobileGUI-RL contains two key components. It (i) synthesizes a curriculum of learnable tasks through self-exploration and filtering, and (ii) adapts GRPO to GUI navigation with trajectory-aware advantages and composite rewards that balance task success and execution efficiency. Experiments on three online mobile-agent benchmarks show consistent gains, validating the effectiveness of our approach.

Figures

Figures reproduced from arXiv: 2507.05720 by the authors.

Figure 1
Figure 1. Framework overall – a scalable pipeline for training GUI agents through self-exploration, [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Training dynamics with and without curriculum learning for 7B and 32B models. The top [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗
Figure 4
Figure 4. Case Studies. The case illustrates the task: “Create a calendar event for tomorrow at 20h [PITH_FULL_IMAGE:figures/full_fig_p017_4.png] view at source ↗

Discussion (0). Sign in to comment.

Forward citations

Cited by 6 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. ReSum: Synergizing LLM Reasoning and Summarization with Reinforcement Learning

    cs.AI 2026-06 unverdicted novelty 7.0 of 10

    ReSum trains LLMs via RLVR to self-summarize reasoning trajectories, yielding 4% average performance gains and 18.6% shorter rollouts through contrastive rollout branches.

  2. SlowBA: An efficiency backdoor attack towards VLM-based GUI agents

    cs.CR 2026-03 conditional novelty 7.0 of 10

    SlowBA uses two-stage reward-level injection and pop-up triggers to make VLM GUI agents produce much longer, slower responses under attack while largely preserving task accuracy.

  3. Scaling GUI Agents with Visual State Transitions

    cs.AI 2026-07 conditional novelty 6.0 of 10

    A joint inverse-forward pretraining stage on visual screen transitions improves GUI-agent fine-tuning by 0.6–6.2 percentage points across three benchmarks.

  4. SCALECUA: Scaling Computer Use Agents with Verifiable Task Synthesis and Efficient Online RL

    cs.AI 2026-07 conditional novelty 6.0 of 10

    Verifiable task synthesis plus frontier sampling and sliding-window visual training yields open-source SOTA computer-use agents on OSWorld and ScienceBoard.

  5. Dark Patterns Meet GUI Agents: LLM Agent Susceptibility to Manipulative Interfaces and the Role of Human Oversight

    cs.HC 2025-09 conditional novelty 6.0 of 10

    GUI agents frequently fall for deceptive interface designs, often without recognizing them, and human supervision of agents improves avoidance only partially while introducing new attention and workload costs.

  6. Parallel-R1: Towards Parallel Thinking via Reinforcement Learning

    cs.CL 2025-09 conditional novelty 6.0 of 10

    Parallel-R1 uses SFT cold-start on easy math plus GRPO on hard math to instill parallel thinking in Qwen3-4B, reporting 8.4% average accuracy gains and a 42.9% AIME25 gain from a parallel-exploration scaffold.

Reference graph

Works this paper leans on

26 extracted references · 3 canonical work pages · cited by 6 Pith papers

  1. [2]

    Anthropic

    Accessed: 2025-06-23. Anthropic. Introducing claude 3.5 sonnet. https://www.anthropic.com/news/claude-3-5-sonnet,

    work page 2025

  2. [3]

    Shuai Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Sibo Song, Kai Dang, Peng Wang, Shijie Wang, Jun Tang, et al. Qwen2. 5-vl technical report. arXiv preprint arXiv:2502.13923,

    arXiv

  3. [5]

    Group-in-group policy optimization for llm agent training

    Lang Feng, Zhenghai Xue, Tingcong Liu, and Bo An. Group-in-group policy optimization for llm agent training. arXiv preprint arXiv:2505.10978,

    arXiv

  4. [6]

    Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning

    Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Ruoyu Zhang, Runxin Xu, Qihao Zhu, Shirong Ma, Peiyi Wang, Xiao Bi, et al. Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning. arXiv preprint arXiv:2501.12948,

    arXiv

  5. [7]

    A real-world webagent with planning, long context understanding, and program synthesis

    11 Technical Report Izzeddin Gur, Hiroki Furuta, Austin Huang, Mustafa Safdari, Yutaka Matsuo, Douglas Eck, and Aleksandra Faust. A real-world webagent with planning, long context understanding, and program synthesis. arXiv preprint arXiv:2307.12856,

    arXiv

  6. [8]

    Webcot: Enhancing web agent reasoning by recon- structing chain-of-thought in reflection, branching, and rollback

    Minda Hu, Tianqing Fang, Jianshu Zhang, Junyu Ma, Zhisong Zhang, Jingyan Zhou, Hongming Zhang, Haitao Mi, Dong Yu, and Irwin King. Webcot: Enhancing web agent reasoning by recon- structing chain-of-thought in reflection, branching, and rollback. arXiv preprint arXiv:2505.20013,

    arXiv

  7. [9]

    Understanding the planning of llm agents: A survey

    Xu Huang, Weiwen Liu, Xiaolong Chen, Xingmei Wang, Hao Wang, Defu Lian, Yasheng Wang, Ruiming Tang, and Enhong Chen. Understanding the planning of llm agents: A survey. arXiv preprint arXiv:2402.02716,

    arXiv

  8. [10]

    Gpt-4o system card

    Aaron Hurst, Adam Lerer, Adam P Goucher, Adam Perelman, Aditya Ramesh, Aidan Clark, AJ Ostrow, Akila Welihinda, Alan Hayes, Alec Radford, et al. Gpt-4o system card. arXiv preprint arXiv:2410.21276,

    arXiv

Show all 26 references
  1. [11]

    Infigui-r1: Advancing multimodal gui agents from reactive actors to deliberative reasoners

    Yuhang Liu, Pengxiang Li, Congkai Xie, Xavier Hu, Xiaotian Han, Shengyu Zhang, Hongxia Yang, and Fei Wu. Infigui-r1: Advancing multimodal gui agents from reactive actors to deliberative reasoners. arXiv preprint arXiv:2504.14239,

    arXiv

  2. [12]

    Arpo: End-to-end policy opti- mization for gui agents with experience replay

    Fanbin Lu, Zhisheng Zhong, Shu Liu, Chi-Wing Fu, and Jiaya Jia. Arpo: End-to-end policy opti- mization for gui agents with experience replay. arXiv preprint arXiv:2505.16282, 2025a. Zhengxi Lu, Yuxiang Chai, Yaxuan Guo, Xi Yin, Liang Liu, Hao Wang, Han Xiao, Shuai Ren, Guanjin...

    arXiv

  3. [13]

    Ui-tars: Pioneering automated gui interaction with native agents

    Yujia Qin, Yining Ye, Junjie Fang, Haoming Wang, Shihao Liang, Shizuo Tian, Junda Zhang, Jiahao Li, Yunxin Li, Shijue Huang, et al. Ui-tars: Pioneering automated gui interaction with native agents. arXiv preprint arXiv:2501.12326,

    arXiv

  4. [14]

    Androidworld: A dynamic benchmarking environment for autonomous agents

    Christopher Rawles, Sarah Clinckemaillie, Yifan Chang, Jonathan Waltz, Gabrielle Lau, Marybeth Fair, Alice Li, William Bishop, Wei Li, Folawiyo Campbell-Ajala, et al. Androidworld: A dynamic benchmarking environment for autonomous agents. arXiv preprint arXiv:2405.14573,

    arXiv

  5. [15]

    Proximal policy optimization algorithms

    John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov. Proximal policy optimization algorithms. arXiv preprint arXiv:1707.06347,

    arXiv

  6. [17]

    Os-genesis: Automating gui agent trajectory construction via reverse task synthesis

    Qiushi Sun, Kanzhi Cheng, Zichen Ding, Chuanyang Jin, Yian Wang, Fangzhi Xu, Zhenyu Wu, Chengyou Jia, Liheng Chen, Zhoumianze Liu, et al. Os-genesis: Automating gui agent trajectory construction via reverse task synthesis. arXiv preprint arXiv:2412.19723,

    arXiv

  7. [18]

    Gui agents with foundation models: A comprehensive survey

    12 Technical Report Shuai Wang, Weiwen Liu, Jingxuan Chen, Yuqi Zhou, Weinan Gan, Xingshan Zeng, Yuhan Che, Shuai Yu, Xinlong Hao, Kun Shao, et al. Gui agents with foundation models: A comprehensive survey. arXiv preprint arXiv:2411.04890,

    arXiv

  8. [19]

    net/forum?id=LPG8pPSfQD

    URL https://openreview. net/forum?id=LPG8pPSfQD. Zhepei Wei, Wenlin Yao, Yao Liu, Weizhi Zhang, Qin Lu, Liang Qiu, Changlong Yu, Puyang Xu, Chao Zhang, Bing Yin, et al. Webagent-r1: Training web agents via end-to-end multi-turn reinforcement learning. arXiv preprint arXiv:2505.16421,

  9. [20]

    Os-atlas: A foundation action model for generalist gui agents

    Zhiyong Wu, Zhenyu Wu, Fangzhi Xu, Yian Wang, Qiushi Sun, Chengyou Jia, Kanzhi Cheng, Zichen Ding, Liheng Chen, Paul Pu Liang, et al. Os-atlas: A foundation action model for generalist gui agents. arXiv preprint arXiv:2410.23218,

    arXiv

  10. [21]

    Zerogui: Automating online gui learning at zero human cost

    Chenyu Yang, Shiqian Su, Shi Liu, Xuan Dong, Yue Yu, Weijie Su, Xuehui Wang, Zhaoyang Liu, Jinguo Zhu, Hao Li, et al. Zerogui: Automating online gui learning at zero human cost. arXiv preprint arXiv:2505.23762,

    arXiv

  11. [22]

    Dapo: An open-source llm reinforcement learning system at scale

    Qiying Yu, Zheng Zhang, Ruofei Zhu, Yufeng Yuan, Xiaochen Zuo, Yu Yue, Weinan Dai, Tiantian Fan, Gaohong Liu, Lingjun Liu, et al. Dapo: An open-source llm reinforcement learning system at scale. arXiv preprint arXiv:2503.14476,

    arXiv

  12. [23]

    Large language model-brained gui agents: A survey

    Chaoyun Zhang, Shilin He, Jiaxu Qian, Bowen Li, Liqun Li, Si Qin, Yu Kang, Minghua Ma, Guyue Liu, Qingwei Lin, et al. Large language model-brained gui agents: A survey. arXiv preprint arXiv:2411.18279, 2024a. Jiwen Zhang, Jihao Wu, Yihua Teng, Minghui Liao, Nuo Xu, Xiao Xiao, ...

    arXiv

  13. [24]

    Gpt-4v (ision) is a generalist web agent, if grounded

    Boyuan Zheng, Boyu Gou, Jihyung Kil, Huan Sun, and Yu Su. Gpt-4v (ision) is a generalist web agent, if grounded. arXiv preprint arXiv:2401.01614,

    arXiv

  14. [25]

    Gui-g1: Understanding r1-zero-like training for visual grounding in gui agents

    Yuqi Zhou, Sunhao Dai, Shuai Wang, Kaiwen Zhou, Qinqlin Jia, et al. Gui-g1: Understanding r1-zero-like training for visual grounding in gui agents. arXiv preprint arXiv:2505.15810,

    arXiv

  15. [26]

    Create a calendar event for tomorrow at 20h with the title ’Call with the Team’ and the description ’We will prepare for team roles.’. The event should last for 30 mins

    13 Technical Report A Detailed Training Configuration A.1 Environment Setup A.1.1 Android Emulator Configuration • Base A VD Name: AndroidWorldAvd • Emulator Instances: Dynamically scaled based on batch size • Screen Resolution: 1080×2400 pixels • Memory Allocation: 3072 MB pe...

    2024

  16. [2017]

    Deepseekmath: Pushing the limits of mathematical reasoning in open language models

    Zhihong Shao, Peiyi Wang, Qihao Zhu, Runxin Xu, Junxiao Song, Xiao Bi, Haowei Zhang, Mingchuan Zhang, YK Li, Y Wu, et al. Deepseekmath: Pushing the limits of mathematical reasoning in open language models. arXiv preprint arXiv:2402.03300,

    arXiv

  17. [2024]

    Anthropic

    Accessed: 2025-06-23. Anthropic. Introducing computer use, a new claude 3.5 sonnet, and claude 3.5 haiku. https: //www.anthropic.com/news/3-5-models-and-computer-use ,

    2025

  18. [2025]

    Webevolver: Enhancing web agent self-improvement with coevolving world model

    Tianqing Fang, Hongming Zhang, Zhisong Zhang, Kaixin Ma, Wenhao Yu, Haitao Mi, and Dong Yu. Webevolver: Enhancing web agent self-improvement with coevolving world model. arXiv preprint arXiv:2504.21024,

    arXiv

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.