Long-horizon GUI agents are a key step toward real-world deployment, yet effective interaction memory under prevailing paradigms remains under-explored. Replaying full interaction sequences is redundant and amplifies noise, while summaries often erase dependency-critical information and traceability. We present AndroTMem, a diagnostic framework for anchored memory in long-horizon Android GUI agents. Its core benchmark, AndroTMem-Bench, comprises 1,069 tasks with 34,473 interaction steps (avg. 32.1 per task, max. 65). We evaluate agents with TCR (Task Complete Rate), focusing on tasks whose completion requires carrying forward critical intermediate state; AndroTMem-Bench is designed to enforce strong step-to-step causal dependencies, making sparse yet essential intermediate states decisive for downstream actions and centering interaction memory in evaluation. Across open- and closed-source GUI agents, we observe a consistent pattern: as interaction sequences grow longer, performance drops are driven mainly by within-task memory failures, not isolated perception errors or local action mistakes. Guided by this diagnosis, we propose Anchored State Memory (ASM), which represents interaction sequences as a compact set of causally linked intermediate-state anchors to enable subgoal-targeted retrieval and attribution-aware decision making. Across multiple settings and 12 evaluated GUI agents, ASM consistently outperforms full-sequence replay and summary-based baselines, improving TCR by 5%-30.16% and AMS by 4.93%-24.66%, indicating that anchored, structured memory effectively mitigates the interaction-memory bottleneck in long-horizon GUI tasks. The code, benchmark, and related resources are publicly available at https://github.com/CVC2233/AndroTMem.
Figures & tables
Figure 1: Overview of AndroTMem . AndroTMem comprises: (i) AndroTMem-Bench , a long-horizon Android GUI benchmark constructed with intent-driven, cross-app, causally dependent workflows; (ii) representative task cases where sparse intermediate states determine downstream decisions; (iii) a diagnostic evaluation suite showing that performance drops with horizon length are primarily caused by memory failures; and (iv) Anchored State Memory (ASM) , which stores causally linked intermediate-state anchors for targeted retrieval and improves long-horizon GUI-agent performance.
Dataset
#Traj
#Steps
#App
GT
Low-Level Instr.
History Modeling
Summary
ASM
AMEX Chai et al. [2025]
2,946
12.8
110
✓
✗
✗
✗
AITW Rawles et al. [2023]
30,378
6.5
357
✓
✗
✗
✗
AITZ Zhang et al. [2024b]
2,504
7.5
70
✓
✓
✓
✗
Android Control Li et al. [2024]
15,283
5.5
833
✓
✓
✓
✗
Android World Rawles et al. [2024]
116
–
20
✗
✗
✗
✗
Table 1: Comparison of AndroTMem-Bench with other mobile GUI agent benchmarks. #Traj: number of trajectories (tasks); #Steps: average interaction steps per trajectory; #App: number of applications. GT: availability of ground-truth trajectories; Low-Level Instr.: low-level instructions. History Modeling: supported within-task history representations.
Figure 2: Overview of the AndroTMem-Bench dataset construction pipeline. (1) Collect popular mobile apps and group them by function. (2) Generate long-horizon cross-app task instructions with step-to-step causal dependencies using dependency-aware templates. (3) Execute and annotate tasks on Android devices or emulators, producing quality-checked trajectories and dataset outputs.
Figure 3: Overview statistics of AndroTMem-Bench. The first row reports the top app combinations, step length distribution by task type, and overall trajectory length distribution (with comparison to prior benchmarks). The second row shows app usage frequency, overall action-type proportions, and action diversity per task.
Model
AMS
TCR
Closed-Source
GPT-4o Hurst et al. [2024]
14.24
11.75
GPT-5 Singh et al. [2025]
12.37
11.46
Gemini-2.5-Flash Comanici et al. [2025]
15.62
14.51
Gemini-2.5-Pro Comanici et al. [2025]
32.71
41.11
Gemini-3-Flash
46.14
55.21
Table 2: Overall performance of evaluated GUI agents on AndroTMem-Bench. AMS measures step-level action accuracy, while TCR evaluates task completion based on anchor-defined intermediate states.
Figure 4: Agent performance across different interaction step ranges under three history utilization strategies: (a) Raw History, (b) Coarse Summary, and (c) Anchored State Memory (ASM).
Figure 5: Model performance across different task categories under three history utilization strategies.
Figure 6: History-related failure modes in long-horizon GUI tasks. Raw traces and coarse summaries often lose task-relevant states, leading to errors such as state loss, mis-binding, context drift, and unverified progress, while ASM preserves structured intermediate states for correct decisions.
Appendix figures & tables6 assets
Supplementary material from the paper’s appendix.
Appendix
Category
Applications
Description
Shopping
JD, Taobao, Pinduoduo, Suning
Product browsing, price comparison, order placement, and online payment for consumer goods.
Travel
Ctrip, Didi Chuxing, Fliggy
Trip planning, transportation booking, ride-hailing services, and real-time travel management.
Map
Amap, Baidu Maps, Tencent Maps
Location search, route planning, navigation, and real-time traffic information.
Mail
Gmail, Outlook
Email communication, message management, information exchange, and productivity-oriented correspondence handling.
Cloud Storage
Baidu Netdisk, Google Drive
Cloud-based file storage, synchronization, backup, and cross-device file sharing.
Meeting
DingTalk, Tencent Meeting, WeCom, Zoom
Online meetings, video conferencing, real-time communication, and collaborative work coordination.
Appendix
Table 5: Functional app grouping based on predefined task-oriented categories (Part I).
Task Type
Primary Intent
Example Scenarios
Lookup
Information retrieval/verification
Checking shop hours, searching for news, or finding answers to specific questions.
Compare & Decide
Decision making via comparison
Comparing menus of three restaurants or checking product parameters across different platforms.
Purchase / Order
Transaction completion
Placing food delivery orders, clearing a shopping cart, or purchasing memberships.
Booking
Resource/Time reservation
Booking hotel rooms, reserving train tickets, or scheduling a doctor’s appointment.
Communicate
Coordination with others
Confirming meeting details with a contact or initiating a group discussion.
Share / Recommend
Content distribution
Sending a store link to a friend or sharing news and music playlists to social media.
Appendix
Table 6: Classification of tasks in AndroTMem based on primary user intent.
Action
Argument
Functionality
TAP
[pos1]
click the on-screen position
INPUT_TEXT
[text]
input the text into the currently focused text field using the system keyboard
LONG_PRESS
[pos1]
press the screen for a long time to copy texts or download images
SWIPE
[pos1]
swipe the screen in a given direction (up, down, left, or right) to scroll the page
SWIPE_TWO_POINT
[pos1,pos2]
swipe from position pos1 to position pos2, enabling precise scrolling or drag-and-drop operations
WAIT
–
wait for a specified duration or until the UI state changes
Appendix
Table 7: The argument and functionality of different actions in AndroTMem. ’pos1’ and ’pos2’ denote the position (x,y) .
Type Name
Recommended Values
Description
Key Subgoal Achievement
SUBGOAL
Completion of intermediate tasks, such as successful login, retrieval of search results, or adding an item to the shopping cart
State Transition and Mode Switching
STATE_CHANGE
Changes in the environment or application mode, such as entering the payment interface or opening a permission request page
Causally Dependent Steps
DEPENDENCY
Steps whose execution depends on prior data or states, such as copying text or filling in a form
Exception Handling
EXCEPTION
Actions taken to address interruptions or errors, such as closing pop-up advertisements or retrying a network request
Global Constraint Information
CONTEXT_INFO
Essential contextual information that constrains task execution, such as selecting dates or setting filter conditions
Task Completion Indicator
FINISH
The sign that the instruction has been completed
Appendix
Table 8: Causal state anchors representing key task progress and execution states in GUI agent interactions.
Figure 7: A complete trajectory example from AndroTMem.
Figure 8: A complete trajectory example from AndroTMem.