Android Bench
AI-assisted software engineering has seen the emergence of several benchmarks to measure the capabilities of LLMs. Android developers face specific challenges that aren't covered by existing benchmarks, so we created one that focuses on a north star of high quality Android development.
Android LLM Leaderboard
| Model | Score (%) |
arrow_range
Cl range (%)
|
Date |
|---|---|---|---|
|
72.4 | 65.1 — 79.3 | 2026-03-16 |
|
72.4 | 64.8 — 79.3 | 2026-02-27 |
|
|
67.7 | 60.1 — 74.8 | 2026-03-18 |
|
66.6 | 58.5 — 74.0 | 2026-02-26 |
|
|
62.5 | 54.8 — 69.8 | 2026-02-26 |
|
|
61.9 | 53.8 — 70.3 | 2026-02-26 |
|
|
60.4 | 52.4 — 68.1 | 2026-02-27 |
|
|
58.4 | 50.9 — 66.5 | 2026-02-27 |
|
|
54.2 | 46.0 — 62.1 | 2026-02-26 |
|
|
42.0 | 36.4 — 47.7 | 2026-02-26 |
|
|
16.1 | 11.2 — 21.2 | 2026-02-26 |
Latest results as of April 7th 2026: This refresh includes the addition of GPT-5.4 and GPT-5.3-Codex.
Check back periodically for updates!
Score is the average percentage of 100 test cases successfully resolved across 10 runs for each model.
Confidence Interval (CI) represents the expected performance range, reflecting the results' statistical reliability (p-value < 0.05).
Check back periodically for updates!
Score is the average percentage of 100 test cases successfully resolved across 10 runs for each model.
Confidence Interval (CI) represents the expected performance range, reflecting the results' statistical reliability (p-value < 0.05).
Learn more about Android Bench
Our methodology
Learn more about how we created a set of common Android developer tasks.
Android best practices
Many of the tasks are based on how we define high quality Android development, which is detailed in our developer documentation.
GitHub repo
See the full repo so you can replicate the tests yourself.