AndroidWorld
在 Android 环境中执行真实应用任务的动态评测环境，按单次尝试的任务成功率（pass@1）排名；榜单为社区提交。
收录成绩4747 个模型或系统
当前第一100.0agent-qa by Vostride AI（gpt-6-sol）
指标任务成功率（pass@1）单次尝试成功完成的任务占比（%）
最近同步10-16 16:571 个数据源 · 任务执行
排名表
显示 47 / 47
| 模型 | 机构 | 参数 | 开源 | 评测设置 | 来源 | 备注 | |||
|---|---|---|---|---|---|---|---|---|---|
| 1 | agent-qa by Vostride AI（gpt-6-sol） | — | — | — | 100.0 | Screenshot + A11y tree | 模型方自报 | 2026-10-16 | 类型：AI agent；代码开源：是；Note: Open-source self-improving QA agent with memory. Code is available at https://github.com/vostride/agent-qa. agent-qa's memory and self-improvement loop were switched off during the benchmark run. |
| 1 | AGI-0（AGI-AW） | — | — | — | 100.0 | Screenshot | 模型方自报 | 2026-10-16 | 类型：AI agent；代码开源：否；[09/2026]: 97.4 --> 100. Trained open source models using our novel training technique on AndroidWorld tasks. Code and model will be open-sourced upon publication. |
| 1 | FluizAI（gpt-4o / gpt-5.6-sol） | — | — | — | 100.0 | Screenshot + A11y tree | 模型方自报 | 2026-10-16 | 类型：AI agent；代码开源：否；Code is not open-sourced yet. |
| 1 | Kirk Engine（gpt-4o / gpt-5.4） | — | — | — | 100.0 | Screenshot + A11y tree | 模型方自报 | 2026-10-16 | 类型：AI agent；代码开源：否；Code is not open-sourced yet. |
| 1 | MobileRun（GPT6 Astra） | — | — | — | 100.0 | Screenshot + A11y tree | 模型方自报 | 2026-10-16 | 类型：AI agent；代码开源：是；[09/30/2026] 91.4 --> Updated harness and using GPT-6 Astra [10/6/2025]: 78.4 --> 91.4; details here [09/19/2025]: Updated score from 63.0 --> 78.4. For this run, we used GPT-5 for reasoning in combination with Gemini-2.5-Pro for acting, while continuing to rely on a hybrid metho… |
| 6 | Artemis（Gemini 3.7 Flash, Gemini Robotics ER 2） | — | — | — | 99.1 | Screenshot + A11y tree | 模型方自报 | 2026-10-16 | 类型：AI agent；代码开源：是；Open-source project available at https://github.com/google/artemis |
| 7 | Finalrun（Gemini 3 flash, Gemini 3 flash lite） | — | — | — | 97.4 | Screenshot + A11y tree | 模型方自报 | 2026-10-16 | 类型：AI agent；代码开源：是；04/2026: Project is open-source at https://github.com/final-run/finalrun-agent 03/2026: Updated results: https://blogs.finalrun.app/finalrun-achieves-the-new-1-sota-mobile-agent-eclipsing-benchmarks-at-97-4 08/2025: The technical doc/approach can be found here: https://github.com… |
| 7 | MobileUseAgent（Seed1.8-GUI） | — | — | — | 97.4 | Screenshot | 模型方自报 | 2026-10-16 | 类型：AI agent；代码开源：是 |
| 9 | askui AndroidVisionAgent, Claude 4.5 Sonnet + Claude 4.0 Sonnet | — | — | — | 94.8 | Screenshot | 模型方自报 | 2026-10-16 | 类型：AI agent；代码开源：是 |
| 9 | AutoDevice（gemini 3 pro + sonnet 4.5） | — | — | — | 94.8 | Screenshot | 模型方自报 | 2026-10-16 | 类型：AI agent；代码开源：是；Agent code is available on https://github.com/autodevice/android_world |
| 11 | Midscene.js（Gemini-3.5-Flash） | — | — | — | 93.1 | Screenshot | 模型方自报 | 2026-10-16 | 类型：AI agent；代码开源：是 |
| 12 | mobile-use（Llama 4-scout, Gemini 2.5 pro, GPT-5 nano） | — | — | — | 91.4 | Screenshot + A11y tree | 模型方自报 | 2026-10-16 | 类型：AI agent；代码开源：是；[12/16/2025] 84.5% -> 91.4% [10/1/2025]: 77.6% -> 84.5% [09/12/2025]: Updated score from 74.1% --> 77.6% [08/19/2025]: Initial trajectory labeling issue (MarkorCreateNoteFromClipboard) has been corrected by the authors. They provide Discord support for setup issues. [08/18/2025]:… |
| 13 | Agent-Visco | — | — | — | 88.8 | Screenshot | 模型方自报 | 2026-10-16 | 类型：AI agent；代码开源：否；Research paper and codes will be open-source in early months. |
| 14 | Surfer 2（o3 + holo1.5-72b） | — | — | — | 87.1 | Screenshot | 模型方自报 | 2026-10-16 | 类型：AI agent；代码开源：否；Agent is based on https://github.com/hcompai/surfer-h-cli - same architecture, but with modifications to prompt and action space |
| 15 | gbox.ai（Sonnet 4.5 + Sonnet 4） | — | — | — | 86.2 | Screenshot | 模型方自报 | 2026-10-16 | 类型：AI agent；代码开源：是；GBOX uses Claude code as the agent with GBOX mcp. Link to report -> https://github.com/babelcloud/android_world_benchmark/blob/main/GBOX/report.md |
| 16 | AutoGLM-Mobile | — | 9B | — | 80.2 | Screenshot + A11y tree | 模型方自报 | 2026-10-16 | 类型：Model；代码开源：否；[9/26/2025]: Updated score from 75.8 --> 80.2 This is an updated version of AutoGLM-Mobile. In this version, we added more pretraining data and trained the VLM at full image resolution. |
| 17 | LX-GUIAgent | — | — | — | 79.3 | Screenshot + A11y tree | 模型方自报 | 2026-10-16 | 类型：AI agent；代码开源：否；[9/23/2025]: Updated score from 75.0 --> 79.3. Added trajectories. |
| 18 | AgentProg（Gemini-2.5-Pro+UI-TARS-1.5） | — | — | — | 78.0 | Screenshot | 模型方自报 | 2026-10-16 | 类型：AI agent；代码开源：是；Paper: https://arxiv.org/pdf/2512.10371; Code: https://github.com/MobileLLM/AgentProg |
| 19 | K²-Agent（Qwen2.5-VL-72B + Qwen2.5-VL-7B） | — | 72B + 7B | — | 76.7 | Screenshot | 模型方自报 | 2026-10-16 | 类型：AI agent；代码开源：否；K²-Agent, a hierarchical approach that self-evolves a Qwen2.5-VL-72B for high-level planning and post-trains a Qwen2.5-VL-7B for low-level execution. Our GitHub repository, which contains a detailed description of our approach, can be found here: https://github.com/k2-agent/k2-ag… |
| 19 | MAI-UI-235B-A22B | — | 235B | — | 76.7 | Screenshot | 模型方自报 | 2026-10-16 | 类型：Model；代码开源：否 |
| 21 | MobileUse-v2（Hammer-UI-32B） | — | 32B | — | 75.0 | Screenshot | 模型方自报 | 2026-10-16 | 类型：AI agent；代码开源：是；Make further post-training based on the GUI-Owl-32b model. Optimize the memory and knowledge module of the MobileUse framework. |
| 22 | MAI-UI-32B | — | 32B | — | 73.3 | Screenshot | 模型方自报 | 2026-10-16 | 类型：Model；代码开源：否 |
| 22 | Mobile-Agent-v3（GUI-Owl-32B） | — | 32B | — | 73.3 | Screenshot | 模型方自报 | 2026-10-16 | 类型：AI agent；代码开源：否；https://github.com/X-PLUG/MobileAgent/tree/main |
| 24 | MAI-UI-8B | — | 8B | — | 70.7 | Screenshot | 模型方自报 | 2026-10-16 | 类型：Model；代码开源：是 |
| 25 | Gemini 2.5 Computer Use | — | — | — | 69.7 | Screenshot | 模型方自报 | 2026-10-16 | 类型：Model；代码开源：否 |
| 26 | JT-GUIAgent-V2 | — | — | — | 67.2 | Screenshot | 模型方自报 | 2026-10-16 | 类型：AI agent；代码开源：否 |
| 27 | GUI-Owl-7B | — | 7B | — | 66.4 | Screenshot | 模型方自报 | 2026-10-16 | 类型：Model；代码开源：否；https://github.com/X-PLUG/MobileAgent/tree/main |
| 28 | UI-Venus-Navi-72B | — | 72B | — | 65.9 | Screenshot | 模型方自报 | 2026-10-16 | 类型：Model；代码开源：是；https://huggingface.co/inclusionAI/UI-Venus-Navi-72B |
| 29 | MobileUse（Qwen2.5-VL-72B） | — | 72B | — | 62.9 | Screenshot | 模型方自报 | 2026-10-16 | 类型：AI agent；代码开源：是 |
| 30 | Seed1.5-VL | — | 20.B | — | 62.1 | Screenshot + A11y tree | 模型方自报 | 2026-10-16 | 类型：Model；代码开源：是 |
| 31 | JT-GUIAgent-V1 | — | — | — | 60.0 | Screenshot | 模型方自报 | 2026-10-16 | 类型：AI agent；代码开源：否 |
| 32 | V-Droid Paper（V-Droid (Llama8B)） | — | 8B | — | 59.5 | A11y tree | 模型方自报 | 2026-10-16 | 类型：AI agent；代码开源：是；Training data consists of apps and tasks from the AndroidWorld benchmark. Code. |
| 33 | Agent S2 | — | — | — | 54.3 | Screenshot | 模型方自报 | 2026-10-16 | 类型：AI agent；代码开源：是 |
| 34 | MAI-UI-2B | — | 2B | — | 49.1 | Screenshot | 模型方自报 | 2026-10-16 | 类型：Model；代码开源：是 |
| 34 | UI-Venus（Venus-Navi-7B） | — | 7B | — | 49.1 | Screenshot | 模型方自报 | 2026-10-16 | 类型：Model；代码开源：是；https://huggingface.co/inclusionAI/UI-Venus-Navi-7B |
| 36 | GUI-Explorer（GPT-4o） | — | — | — | 47.4 | Screenshot + A11y tree | 模型方自报 | 2026-10-16 | 类型：AI agent；代码开源：是 |
| 37 | AndroidGen（GPT-4o） | — | — | — | 46.8 | A11y tree | 模型方自报 | 2026-10-16 | 类型：AI agent；代码开源：是 |
| 38 | UI-TARS | — | 72B | — | 46.6 | Screenshot | 模型方自报 | 2026-10-16 | 类型：Model；代码开源：是 |
| 39 | GPT-4o + Aria-UI | — | — | — | 44.8 | Screenshot | 模型方自报 | 2026-10-16 | 类型：Model；代码开源：是 |
| 40 | GPT-4o + UGround | — | — | — | 44.0 | Screenshot | 模型方自报 | 2026-10-16 | 类型：Model；代码开源：是；Code for reproduction |
| 40 | ScaleTrack-7B | — | 8B | — | 44.0 | A11y tree | 模型方自报 | 2026-10-16 | 类型：Model；代码开源：否 |
| 42 | Mirage-1（GPT-4o） | — | — | — | 42.2 | Screenshot | 模型方自报 | 2026-10-16 | 类型：AI agent；代码开源：是；With Mirage-1-O; uses OS-Atlas grounder |
| 43 | Ponder & Press（GPT-4o） | — | — | — | 34.5 | Screenshot | 模型方自报 | 2026-10-16 | 类型：AI agent；代码开源：否；Code is not yet open-sourced |
| 44 | AndroidWorld（GPT-4 Turbo） | — | — | — | 30.6 | A11y tree | 模型方自报 | 2026-10-16 | 类型：AI agent；代码开源：是 |
| 45 | EcoAgent（GPT-4o, OS-Atlas-Pro 4B, Qwen2-VL-2B-Instruct） | — | — | — | 27.6 | Screenshot | 模型方自报 | 2026-10-16 | 类型：AI agent；代码开源：否 |
| 45 | GUI-Critic-R1（Qwen-2.5-VL-7B） | — | 7B | — | 27.6 | Screenshot + A11y tree | 模型方自报 | 2026-10-16 | 类型：Model；代码开源：是 |
| 47 | InfiGUIAgent（Qwen2-VL-2B (fine-tuned)） | — | 2B | — | 9.0 | Screenshot | 模型方自报 | 2026-10-16 | 类型：Model；代码开源：是 |
榜单说明
第 1 批接入- 指标
- 任务成功率（pass@1）：单次尝试成功完成的任务占比（%）（越高越好）
- 数据源
- 来源类型
- 模型方自报 47
分数取自数据源原文；同一模型在不同评测设置或不同来源下分别列出，来源冲突时全部保留并标注来源类型。
更新记录
0 条首次同步作为基线，之后新上榜、分数变化和移除的条目会记录在这里。