Skip to content

Benchmark

AndroidWorld agent system

Agentic benchmarks measure how well models carry out tasks that take several steps. Scores are percentages. Higher is better.

Models and systems measured
14
Scores recorded
14
Sources
1
IQ weight
No weight published

Score distribution

71.6%
Mean
75%
Median
26.1%
Std dev
34.5%
10th percentile
100%
90th percentile
30.6%
Min
30.6%100%
The band covers the middle 80% of scores. The line marks the median.

Results

The source marks these results as descriptive, not capability scores, so we list them by name instead of ranking them.

Model or systemScore
agent-qa by Vostride AIgpt-6-sol, Screenshot + A11y tree, 1· AndroidWorld leaderboard data docs100%
AGI-0AGI-AW, Screenshot, 1· AndroidWorld leaderboard data docs100%
AndroidGenGPT-4o, A11y tree, 1· AndroidWorld leaderboard data docs46.8%
AndroidWorldGPT-4 Turbo, A11y tree, 1· AndroidWorld leaderboard data docs30.6%
GUI-ExplorerGPT-4o, Screenshot + A11y tree, 1· AndroidWorld leaderboard data docs47.4%
Midscene.jsGemini-3.5-Flash, Screenshot, 1· AndroidWorld leaderboard data docs93.1%
Mirage-1GPT-4o, Screenshot, 1· AndroidWorld leaderboard data docs42.2%
Mobile-Agent-v3GUI-Owl-32B, Screenshot, 1· AndroidWorld leaderboard data docs73.3%
MobileRunGPT6 Astra, Screenshot + A11y tree, 1· AndroidWorld leaderboard data docs100%
MobileUseQwen2.5-VL-72B, Screenshot, 1· AndroidWorld leaderboard data docs62.9%
MobileUse-v2Hammer-UI-32B, Screenshot, 1· AndroidWorld leaderboard data docs75%
MobileUseAgentSeed1.8-GUI, Screenshot, 1· AndroidWorld leaderboard data docs97.4%
Ponder & PressGPT-4o, Screenshot, 1· AndroidWorld leaderboard data docs34.5%
QalionClaude Haiku 5.5, Screenshot + A11y tree, 1· AndroidWorld leaderboard data docs99.1%

Data credits: AndroidWorld leaderboard sheet, Google Research android_world. Community-submitted, self-reported results without independent verification; sheet licence not stated (code Apache-2.0 not applied).