AndroidWorld

Google Research environment and benchmark for autonomous Android control agents: 116 hand-crafted tasks across 20 apps on a live emulator.

Category

Information

Website

https://google-research.github.io/android_world/

Repository

https://github.com/google-research/android_world

Docs

https://github.com/google-research/android_world#readme

Status

active

Pricing

open-source

License

Apache-2.0

Maintainer

Google Research

Added

2026-10-25

Updated

2026-10-25

Last verified

2026-10-25

Submitted by

agentica-curator

Tags

• benchmark
• mobile
• android
• computer-use

Description

Tasks are instantiated with random parameters, giving millions of unique variations with durable reward signals; also supports MiniWoB++ web tasks. Runs against an Android emulator (Pixel 6, API 33) with ~2 GB memory and 8 GB disk; experimental Docker support. Links to a public leaderboard spreadsheet from the project site.

Related

• OSWorld: NeurIPS 2024 benchmark of multimodal agents on open-ended tasks in real computer environments.
• Windows Agent Arena: Microsoft's scalable OS benchmark for multimodal agents that perform tasks inside a real Windows environment, parallelizable on Azure.

Sources

• https://github.com/google-research/android_world
• https://google-research.github.io/android_world/
• https://arxiv.org/pdf/2405.14573

Machine-readable

• JSON
• Markdown
• Source file on GitHub (edit via pull request)