Pith. sign in

REVIEW 4 major objections 4 minor 1 cited by

A roughly 3-billion-parameter mobile agent matches or beats much larger rivals by co-scaling data and environments instead of parameters.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-02 01:46 UTC pith:2FYMRQOO

load-bearing objection Serious engineering, plausible numbers, but the main AndroidWorld result is not independent because the policy trains in that same environment. the 4 major comments →

arxiv 2607.14548 v1 pith:2FYMRQOO submitted 2026-07-16 cs.CV

HyMobileAgent: Data-Environment Co-Scaling for Efficient GUI Agents

Hy Vision Team , Huawen Shen , Zhengyang Tang , Shangpin Peng , Liang Wu , Anran Zhang , Weinong Wang , Yiduo Guo
show 15 more authors
Chenxin Li Zhengyao Fang Yang Ding Junyi Li Fei Tang Zheng Ruan Yi Zhang Xingran Zhou Dingchen Yang Sunqi Fan Zhiyi Wan Han Hu Xin Lai Pengyuan Lyu Chengquan Zhang
This is my paper
classification cs.CV
keywords mobile GUI agentjoint data-environment scalingPhoneWorldAndroidWorldGUI groundingreinforcement learninglong-horizon decision makingmock application environment
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

HyMobileAgent is built on a ~3B-parameter vision-language model and claims that scaling training data and interaction environments, not model size, is what makes a mobile GUI agent succeed. The paper reports 82.6% strict success on AndroidWorld and 42% on its own real-device benchmark HyMobileWorld, matching or exceeding far larger general-purpose agents. It attributes the gains to a closed-loop data system (mock-interface synthesis, tutorial-to-data distillation, million-scale action trajectories), a resettable mock-app environment called PhoneWorld, and a Planning-and-Reflection decision structure with dead-loop detection, trained via mid-training, SFT, and two-phase RL. The broader claim is that, at a fixed deployment budget, environment-grounded reinforcement learning can deliver returns that rival or exceed adding parameters.

Core claim

HyMobileAgent, built on the Hy3.0-VL-A3B vision-language model, achieves an 82.6% strict success rate on AndroidWorld and 42.0% on the in-house HyMobileWorld real-device benchmark, matching or beating substantially larger general-purpose agents while keeping an A3B-scale deployment footprint. The paper's central claim is that the binding constraint for small mobile agents is not foundation-model capacity but the quality of surrounding data, interaction environments, and decision structure. It presents a joint data- and environment-centric scaling framework — a GUI perception flywheel, knowledge extraction from tutorial videos, a million-scale action-trajectory pipeline over 2,000+ sandbox an

What carries the argument

The load-bearing mechanism is the joint data- and environment-scaling loop. On the data side, a perception flywheel combines mock-interface synthesis, reject-sampling difficulty selection, and icon-specific augmentation; a knowledge pipeline turns tutorial videos into single-image planning data and multi-image state-transition data; and an action pipeline collects million-scale trajectories with automated failure attribution. On the environment side, PhoneWorld provides 34 resettable mock apps and over 34,000 verifiable tasks, making trajectory-level reinforcement learning possible without login, payment, or anti-bot barriers. The decision structure that carries long-horizon behavior is the

Load-bearing premise

The weakest load-bearing premise is that PhoneWorld's 34 mock apps and sandbox/real-device pipelines yield supervision that transfers to real mobile applications; the paper provides no controlled ablation measuring mock-to-real transfer, so if that transfer fails, the environment-scaling explanation for the gains collapses.

What would settle it

Train two otherwise identical agents, one with PhoneWorld mock-app data and the other with only real-device trajectories, then compare AndroidWorld and HyMobileWorld success rates. If the mock-only agent loses most of the gap, the mock-to-real transfer assumption is false and the co-scaling claim is weakened.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • At a fixed deployment budget, investing in environment-grounded RL and self-correcting decision structures can close much of the gap to much larger models.
  • Verifiable, resettable mock-app environments make RL training feasible for tasks that real devices block (logins, payments, anti-bot), so environment design becomes a first-class scaling axis.
  • Tutorial videos and books can be distilled into structured planning and state-transition supervision, adding procedural knowledge without new human annotation.
  • The three-repeat dead-loop trigger gives error recovery a learnable, deterministic target, which should reduce compounding-error failures in long trajectories.
  • The large gap between AndroidWorld (82.6%) and HyMobileWorld (42%) suggests real-device, mini-program, and cross-app scenarios remain the hardest frontier for all agents.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The paper compares against other models but does not ablate the co-scaling components individually; a controlled study removing only the mock-app/PhoneWorld portion would test whether environment scaling, rather than the larger SFT corpus, drives the gains.
  • If mock-to-real transfer holds, mobile agents could be trained almost entirely in synthetic resettable environments and fine-tuned on a thin slice of real data, sharply reducing the cost of agent training.
  • Making recovery conditions deterministic and trainable, as in the dead-loop reflection design, could transfer beyond mobile GUIs to web and desktop agents.
  • The in-house benchmarks are described in the report but are not part of the public benchmark set, so independent re-evaluation awaits their release.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. HyMobileAgent is an A3B-scale mobile GUI agent built on Hy3.0-VL-A3B, trained with a joint data- and environment-scaling framework: a GUI perception flywheel, tutorial-based knowledge extraction, a large-scale action trajectory pipeline over sandbox and real devices, the PhoneWorld mock-app environment, a structured Planning-and-Reflection mechanism, and a three-stage training recipe (mid-training, SFT, offline/online RL). The paper reports an 82.6% success rate on AndroidWorld and 42% on an in-house HyMobileWorld benchmark, claiming to match or surpass substantially larger proprietary agents at a much smaller deployment footprint, and concludes that data/environment scaling can rival additional model parameters.

Significance. If independently verified, the central claim is significant: a ~3B-scale mobile agent outperforming or matching much larger general-purpose models on AndroidWorld would support the thesis that environment-grounded data scaling has high marginal return. The paper also contains a concrete decision structure, a deterministic dead-loop reflection mechanism, and a large-scale data pipeline design that are of interest to the GUI-agent community. However, the evidence as presented is not yet load-bearing: AndroidWorld appears in the online-RL training pool, the in-house benchmarks are unreleased, and no ablation isolates the contribution of PhoneWorld or mock-app data. These are not presentation issues; they directly affect whether the headline comparison measures transfer or overfitting to the evaluation environment.

major comments (4)
  1. [§5.3, Table 2; §6.1, Table 3] The headline AndroidWorld result comes from a policy whose online-RL rollout pool includes approximately 500 AndroidWorld devices (Table 2), and AndroidWorld is then used in §6.1 as the primary public action-execution benchmark. No withheld split, per-task overlap analysis, or ablation excluding AndroidWorld from the RL pool is reported. The comparison to Gemini 3.1, Seed 2.0 Pro, GPT-5.4-Pro, and Claude-4.7 Opus is therefore not a like-for-like test of generalization: those systems were not trained on AndroidWorld rewards. This is the central load-bearing issue for the abstract's claim. A revision should either train a variant without AndroidWorld rollouts and report its AndroidWorld score, or provide a strict task-disjoint holdout analysis.
  2. [§4.4, §5.3] PhoneWorld is used for SFT-data generation, as the RL task pool (34,242 single-app and 500 cross-app tasks), and as part of the online-RL environment mixture (34 mock apps, ~1,000 VM instances). Yet PhoneWorld is also a self-cited prior paper (Tang et al., 2026) with overlapping authors. There is no controlled ablation that removes PhoneWorld or mock-app data from training, and no direct measurement of whether PhoneWorld tasks transfer to real Android apps. Without such an ablation, the causal attribution of the reported gains to 'environment co-scaling' is not established; the gains could be dominated by other data sources or by the real-app rollout pools.
  3. [§6.2, Table 3] All action-execution numbers are reported as single point estimates. AndroidWorld is a dynamic environment with stochastic task states, and HyMobileWorld involves human grading, yet no error bars, number of independent runs, per-task standard deviations, inter-annotator agreement, or statistical tests are provided. Differences such as 82.6 vs. 80.2 (AndroidWorld) and 42.0 vs. 44.7 (HyMobileWorld) are within plausible noise for such evaluations. The claims of 'matching or surpassing' are therefore not statistically supported as reported.
  4. [§6.1] HyMobileWorld, HyMobileGrounding, and HyMobileQA are in-house benchmarks that are not released. For HyMobileWorld, the evaluation description is limited to 150 tasks ('50 native-app, 50 mini-program, 50 cross-app') with trained annotators performing blinded assessments; no task list, annotation protocol, release plan, or independent replication is provided. The 42% figure on HyMobileWorld and the in-house grounding/QA numbers therefore cannot be externally verified, which is particularly problematic because these in-house suites are used to support the central transfer claim.
minor comments (4)
  1. [Abstract] The abstract refers to 'HyMobileOnline' but the benchmark is consistently called 'HyMobileWorld' elsewhere in the paper; please unify the name.
  2. [§5.3] The action-reward description states that 'the exact weighting and the per-primitive formulations are deferred to the released training configuration,' but no release URL, repository, or configuration artifact is provided. Please include the configuration or a pointer to a public release.
  3. [§6.2/Figure 1] Figure 1 and Table 3 report scores without any uncertainty or run count. Even if error bars are not possible for all proprietary baselines, the authors' own runs should include variability estimates, and the figure should distinguish published numbers from self-measured ones.
  4. [§3.1] The term 'A3B' is used throughout without a definition. Please specify whether it means 3 billion active parameters in a larger model, 3 billion total parameters, or something else, and clarify how it relates to the 'A3B-scale deployment footprint.'

Circularity Check

1 steps flagged

Headline AndroidWorld result is not independent: AndroidWorld appears in the online-RL rollout pool and is then used as the primary public benchmark.

specific steps
  1. fitted input called prediction [§5.3 (Online RL) / Table 2; §6.1 (Benchmarks) / Table 3]
    "The setup combines roughly 500 AndroidWorld devices, around 1,200 devices for real-app rollouts, and around 1,000 Android virtual-machine instances for mock-app rollouts. ... AndroidWorld has become the most widely adopted public mobile-agent benchmark and is used here as the primary public reference."

    The paper's headline success on AndroidWorld is the main evidence for the claim that environment co-scaling, not model scale, drives mobile-agent gains. But §5.3 places ~500 AndroidWorld devices in the online-RL trajectory sampling pool, so the policy was optimized with rewards from AndroidWorld rollouts. Evaluating on AndroidWorld then measures in-distribution performance, not transfer to an unseen public benchmark. The comparison with Gemini/Seed/GPT/Claude is not like-for-like, because those baselines were not trained on AndroidWorld. No withheld AndroidWorld split or ablation excluding AndroidWorld from the RL pool is reported, so the 'prediction' is not independent of the training input.

full rationale

The central derivation chain is: build data/environment scaling -> train with online RL over a heterogeneous pool -> evaluate on AndroidWorld as the public reference. The paper explicitly lists ~500 AndroidWorld devices in the online-RL sampling pool (§5.3, Table 2) and then designates AndroidWorld as 'the primary public reference' (§6.1). Thus the headline 82.6% is not an out-of-distribution test of the co-scaling recipe; it evaluates the policy in an environment whose reward signal was used during training. No withheld AndroidWorld split and no ablation that removes AndroidWorld from the RL pool are reported. This is a train/evaluation overlap rather than a definitional equivalence, so I score it as partial circularity (6), not as a fully forced result. Other reported axes are more independent: grounding and QA are measured on external public benchmarks (ScreenSpot V2, ScreenSpot-Pro, MMBench-GUI) that are not listed in the training environment pool, and those results are not contaminated. PhoneWorld is self-cited but the current paper describes its role directly, so I do not treat that self-citation as separately load-bearing for the circularity score.

Axiom & Free-Parameter Ledger

4 free parameters · 4 axioms · 2 invented entities

The report's central result rests on self-built infrastructure that is not released: the base model, the mock environment, and the in-house benchmarks. This is not a graviton-style theoretical entity, but it is a practical version of the same problem: the reader cannot check the load-bearing components.

free parameters (4)
  • Dead-loop threshold = 3
    Number of consecutive identical actions that triggers reflection (§3.3); chosen by hand, and the main new mechanism's sensitivity is not ablated.
  • Reject-sampling bucket boundaries = 0/8, 1/7–7/8, 8/8
    Difficulty stratification in perception flywheel (§4.1); arbitrary choice of 8 samples and bucket cut points, not justified by data.
  • Online-RL environment mixture = ~500 AndroidWorld, ~1,200 real-app, ~1,000 VM
    Table 2; chosen by hand, includes AndroidWorld as training environment, directly affecting the headline evaluation.
  • Action reward weighting
    §5.3: 'exact weighting and per-primitive formulations are deferred to the released training configuration'; a free parameter required for the offline RL action reward, unreported.
axioms (4)
  • ad hoc to paper Hy3.0-VL-A3B provides native any-resolution input up to 4K, an A3B parameter budget, and a 32K context window.
    §3.1; the properties are asserted without release or independent verification, and the whole system depends on them.
  • domain assumption PhoneWorld mock-app trajectories transfer to real-device success.
    §4.4/§5.3; no controlled transfer ablation; only in-house HyMobileWorld evidence.
  • domain assumption Baseline scores for proprietary models are comparable.
    Tables 3-5; acquisition protocol not described.
  • domain assumption Automated failure attribution and tutorial distillation produce correct supervision.
    §4.2-4.3; no quality measurements or human validation reported.
invented entities (2)
  • Hy3.0-VL-A3B no independent evidence
    purpose: Vision-language backbone for the whole agent
    Unreleased model; no weights/checkpoints; no independent evaluation.
  • PhoneWorld Mock App Factory no independent evidence
    purpose: Resettable training/evaluation environment with 34 mock apps and 34k tasks
    Introduced via self-cited prior paper; not released or independently verified.

pith-pipeline@v1.3.0-alltime-deepseek · 23114 in / 16501 out tokens · 165623 ms · 2026-08-02T01:46:13.979940+00:00 · methodology

0 comments
read the original abstract

As large multimodal models move from understanding content to operating on digital environments, mobile GUI has emerged as a challenging and consequential testbed for digital embodied intelligence. Mobile agents operate under three coupled constraints: precise perception of complex interfaces, scalable acquisition of high-quality interaction data, and robust long-horizon decision making under compounding execution errors. This report presents HyMobileAgent, a mobile GUI agent built on Hy3.0-VL-A3B, a vision-native foundation model featuring native any-resolution input, an A3B-scale deployment budget, and a 32K context window to model extended interaction histories. Rather than relying solely on model scaling, we develop a joint data and environment centric scaling framework to address the key bottlenecks of mobile interaction. Our framework integrates a GUI perception flywheel combining mock-interface synthesis, rejection sampling, and icon-specific augmentation; a knowledge pipeline that transforms tutorial videos into structured interaction data; a million-scale action data pipeline deployed across more than 2000 sandbox and real-device instances with automated failure attribution; the PhoneWorld Mock App Factory, providing a resettable training environment with 34 mock applications and over 34000 tasks; and a structured Planning-and-Reflection mechanism with explicit dead-loop detection for reliable long-horizon execution. We also introduce a progressive training recipe consisting of mid-training, supervised fine-tuning, and reinforcement learning with task-specific reward designs.

Figures

Figures reproduced from arXiv: 2607.14548 by Anran Zhang, Chengquan Zhang, Chenxin Li, Dingchen Yang, Fei Tang, Han Hu, Huawen Shen, Hy Vision Team, Junyi Li, Liang Wu, Pengyuan Lyu, Shangpin Peng, Sunqi Fan, Weinong Wang, Xingran Zhou, Xin Lai, Yang Ding, Yiduo Guo, Yi Zhang, Zheng Ruan, Zhengyang Tang, Zhengyao Fang, Zhiyi Wan.

Figure 1
Figure 1. Figure 1: Performance comparison of HyMobileAgent and other SOTA models. [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Representative capability profile of HyMobileAgent: (1) interaction with apps and closed [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Overview of the HyMobileAgent agent loop. Starting from the user instruction, HyMobileAgent [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Inference and training template used by HyMobileAgent. The five-field Thought block precedes [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: The HyMobileAgent data system. Under a unified task taxonomy, three layers cover [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Composition of HyMobileWorld. The 150 tasks are evenly split across three categories (50 native [PITH_FULL_IMAGE:figures/full_fig_p016_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Single-app task: querying real-time traffic from Huaqiangbei to Shenzhen North Station in [PITH_FULL_IMAGE:figures/full_fig_p023_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Single-app task: searching for a crosstalk album in Ximalaya, playing the first album at 1.5 [PITH_FULL_IMAGE:figures/full_fig_p024_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Single-app task: applying a retro filter to the first photo in the album using Meitu. [PITH_FULL_IMAGE:figures/full_fig_p025_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Mini-program task: checking HPV vaccine appointment availability in the Tencent Health [PITH_FULL_IMAGE:figures/full_fig_p026_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Mini-program task: searching sunscreen in the VIP.com mini-program and inspecting the [PITH_FULL_IMAGE:figures/full_fig_p027_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Mini-program task: entering origin and destination in the DiDi mini-program and comparing [PITH_FULL_IMAGE:figures/full_fig_p028_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Mini-program task: querying train tickets and filtering by departure time in the Tongcheng [PITH_FULL_IMAGE:figures/full_fig_p029_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Cross-app task: searching a news headline in Tencent News, then checking the same topic’s [PITH_FULL_IMAGE:figures/full_fig_p030_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Cross-app task: finding a neck-relaxation course in Keep, then playing matching white noise [PITH_FULL_IMAGE:figures/full_fig_p031_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Cross-app task: opening a trending costume drama in Tencent Video, then looking up its rating [PITH_FULL_IMAGE:figures/full_fig_p032_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Cross-app task: asking Yuanbao for jazz recommendations, then searching the recommended [PITH_FULL_IMAGE:figures/full_fig_p033_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: Element-level parsing of desktop scenarios. HyMobileAgent provides fine-grained global [PITH_FULL_IMAGE:figures/full_fig_p034_18.png] view at source ↗
Figure 19
Figure 19. Figure 19: Element-level parsing of mobile scenarios. HyMobileAgent provides accurate and fine-grained [PITH_FULL_IMAGE:figures/full_fig_p035_19.png] view at source ↗
Figure 20
Figure 20. Figure 20: Instruction-guided desktop localization. HyMobileAgent interprets the user instruction and [PITH_FULL_IMAGE:figures/full_fig_p036_20.png] view at source ↗
Figure 21
Figure 21. Figure 21: Instruction-guided mobile localization. HyMobileAgent performs fine-grained global local [PITH_FULL_IMAGE:figures/full_fig_p036_21.png] view at source ↗
Figure 22
Figure 22. Figure 22: This example demonstrates the general question-answering capability of HyMobileAgent in [PITH_FULL_IMAGE:figures/full_fig_p037_22.png] view at source ↗
Figure 23
Figure 23. Figure 23: By introducing explicit constraints in the prompt, HyMobileAgent can organize its reasoning [PITH_FULL_IMAGE:figures/full_fig_p038_23.png] view at source ↗
Figure 24
Figure 24. Figure 24: HyMobileAgent also exhibits strong generalization ability in non-GUI visual understanding [PITH_FULL_IMAGE:figures/full_fig_p038_24.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 1 Pith paper

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

  1. Qwen-UI-Agent Technical Report: Toward Next-Generation Real-World Centric Foundation GUI Agents

    cs.AI 2026-07 conditional novelty 5.5

    A real-device-centric foundation GUI agent with hybrid GUI+CLI batched actions, AutoResearch data flywheel, online RL, and a proactive harness reaches SOTA mobile and competitive desktop/web scores.

Reference graph

Works this paper leans on

22 extracted references · 12 linked inside Pith · cited by 1 Pith paper

  1. [6]

    Gui-r1: A generalist r1-style vision-language action model for gui agents.arXiv preprint arXiv:2504.10458,

    Run Luo, Lu Wang, Wanwei He, Longze Chen, Jiaming Li, and Xiaobo Xia. Gui-r1: A generalist r1-style vision-language action model for gui agents.arXiv preprint arXiv:2504.10458,

    Pith/arXiv arXiv

  2. [8]

    Androidworld: A dynamic benchmarking environment for autonomous agents

    Chris 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. InInternational Conference on Learning Representations, volume 2025, pp. 406–441,

    2025

  3. [11]

    Phoneworld: Scaling phone-use agent environments.arXiv preprint arXiv:2605.29486,

    Zhengyang Tang, Yuxuan Liu, Xin Lai, Junyi Li, Pengyuan Lyu, Yiduo Guo, Zhengyao Fang, Yang Ding, Yi Zhang, Weinong Wang, et al. Phoneworld: Scaling phone-use agent environments.arXiv preprint arXiv:2605.29486,

    Pith/arXiv arXiv

  4. [12]

    Ui-venus-1.5 technical report.arXiv preprint arXiv:2602.09082,

    Venus Team, Changlong Gao, Zhangxuan Gu, Yulin Liu, Xinyu Qiu, Shuheng Shen, Yue Wen, Tianyu Xia, Zhenyu Xu, Zhengwen Zeng, et al. Ui-venus-1.5 technical report.arXiv preprint arXiv:2602.09082,

    arXiv

  5. [13]

    UI-TARS-2 technical report: Advancing GUI agent with multi-turn reinforcement learning, 2025a

    Haoming Wang, Yujia Qin, Yining Ye, Shihao Liang, Bo Yu, Sicheng Hu, Junda Zhang, Sheng Yang, Shizuo Tian, Junjie Fang, Pengfei Bi, Tianlin Ye, Tao Lin, and Yang Sun. UI-TARS-2 technical report: Advancing GUI agent with multi-turn reinforcement learning, 2025a. Junyang Wang, Haiyang Xu, Jiabo Ye, Ming Yan, Weizhou Shen, Ji Zhang, Fei Huang, and Jitao Sang...

    Pith/arXiv arXiv

  6. [14]

    Mmbench-gui: Hierarchical multi-platform evaluation framework for gui agents.arXiv preprint arXiv:2507.19478, 2025b

    Xuehui Wang, Zhenyu Wu, JingJing Xie, Zichen Ding, Bowen Yang, Zehao Li, Zhaoyang Liu, Qingyun Li, Xuan Dong, Zhe Chen, et al. Mmbench-gui: Hierarchical multi-platform evaluation framework for gui agents.arXiv preprint arXiv:2507.19478, 2025b. Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, Denny Zhou, et al. Chain-of-t...

    Pith/arXiv arXiv

  7. [15]

    Os-atlas: 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: Foundation action model for generalist gui agents. InInternational Conference on Learning Representations, volume 2025, pp. 5090–5108,

    2025

  8. [16]

    Mobile-agent-v3

    21 Haiyang Xu, Xi Zhang, Haowei Liu, Junyang Wang, Zhaozai Zhu, Shengjie Zhou, Xuhao Hu, Feiyu Gao, Junjie Cao, Zihua Wang, et al. Mobile-agent-v3. 5: Multi-platform fundamental gui agents.arXiv preprint arXiv:2602.16855,

    arXiv

  9. [17]

    Mobilerl: Online agentic reinforcement learning for mobile gui agents

    Yifan Xu, Xiao Liu, Xinghan Liu, Jiaqi Fu, Hanchen Zhang, Bohao Jing, Shudan Zhang, Yuting Wang, Wenyi Zhao, and Yuxiao Dong. Mobilerl: Online agentic reinforcement learning for mobile gui agents. arXiv preprint arXiv:2509.18119, 2025a. Yifan Xu, Xiao Liu, Xueqiao Sun, Siyi Cheng, Hao Yu, Hanyu Lai, Shudan Zhang, Dan Zhang, Jie Tang, and Yuxiao Dong. Andr...

    arXiv

  10. [18]

    Aria-ui: Visual grounding for gui instructions

    Yuhao Yang, Yue Wang, Dongxu Li, Ziyang Luo, Bei Chen, Chao Huang, and Junnan Li. Aria-ui: Visual grounding for gui instructions. InFindings of the Association for Computational Linguistics: ACL 2025, pp. 22418–22433,

    2025

  11. [19]

    React: Synergizing reasoning and acting in language models.arXiv preprint arXiv:2210.03629,

    Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, and Yuan Cao. React: Synergizing reasoning and acting in language models.arXiv preprint arXiv:2210.03629,

    Pith/arXiv arXiv

  12. [20]

    Mobile-agent-v3: Fundamental agents for gui automation.arXiv preprint arXiv:2508.15144,

    Jiabo Ye, Xi Zhang, Haiyang Xu, Haowei Liu, Junyang Wang, Zhaoqing Zhu, Ziwei Zheng, Feiyu Gao, Junjie Cao, Zhengxi Lu, et al. Mobile-agent-v3: Fundamental agents for gui automation.arXiv preprint arXiv:2508.15144,

    Pith/arXiv arXiv

  13. [21]

    Appagent: Multimodal agents as smartphone users

    Chi Zhang, Zhao Yang, Jiaxuan Liu, Yanda Li, Yucheng Han, Xin Chen, Zebiao Huang, Bin Fu, and Gang Yu. Appagent: Multimodal agents as smartphone users. InProceedings of the 2025 CHI Conference on Human Factors in Computing Systems, pp. 1–20, 2025a. Miaosen Zhang, Ziqiang Xu, Jialiang Zhu, Qi Dai, Kai Qiu, Yifan Yang, Chong Luo, Tianyi Chen, Justin Wagle, ...

    Pith/arXiv arXiv 2025

  14. [22]

    Webarena: A realistic web environment for building autonomous agents

    Shuyan Zhou, Frank F Xu, Hao Zhu, Xuhui Zhou, Robert Lo, Abishek Sridhar, Xianyi Cheng, Tianyue Ou, Yonatan Bisk, Daniel Fried, et al. Webarena: A realistic web environment for building autonomous agents. InInternational Conference on Learning Representations, volume 2024, pp. 15585–15606,

    2024

  15. [2014]

    Fara-7b: An efficient agentic model for computer use.arXiv preprint arXiv:2511.19663,

    Ahmed Awadallah, Yash Lara, Raghav Magazine, Hussein Mozannar, Akshay Nambi, Yash Pandya, Aravind Rajeswaran, Corby Rosset, Alexey Taymanov, Vibhav Vineet, et al. Fara-7b: An efficient agentic model for computer use.arXiv preprint arXiv:2511.19663,

    arXiv

  16. [2017]

    Deepseekmath: Pushing the limits of mathematical reasoning in open language models.arXiv preprint arXiv:2402.03300,

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

    Pith/arXiv arXiv

  17. [2020]

    Autoglm: Autonomous foundation agents for guis.arXiv preprint arXiv:2411.00820,

    Xiao Liu, Bo Qin, Dongzhu Liang, Guang Dong, Hanyu Lai, Hanchen Zhang, Hanlin Zhao, Iat Long Iong, Jiadai Sun, Jiaqi Wang, et al. Autoglm: Autonomous foundation agents for guis.arXiv preprint arXiv:2411.00820,

    Pith/arXiv arXiv

  18. [2022]

    Ui-tars: Pioneering automated gui interaction with native agents.arXiv preprint arXiv:2501.12326,

    20 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,

    Pith/arXiv arXiv

  19. [2023]

    Proximal policy optimization algorithms.arXiv preprint arXiv:1707.06347,

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

    Pith/arXiv arXiv

  20. [2024]

    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,

    Pith/arXiv arXiv

  21. [2025]

    Auto-explorer: Automated data collection for gui agent

    Xiangwu Guo, Difei Gao, and Mike Zheng Shou. Auto-explorer: Automated data collection for gui agent. arXiv preprint arXiv:2511.06417,

    arXiv

  22. [2026]

    Mobiledreamer: Generative sketch world model for gui agent.arXiv preprint arXiv:2601.04035,

    Yilin Cao, Yufeng Zhong, Zhixiong Zeng, Liming Zheng, Jing Huang, Haibo Qiu, Peng Shi, Wenji Mao, and Wan Guanglu. Mobiledreamer: Generative sketch world model for gui agent.arXiv preprint arXiv:2601.04035,

    arXiv