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 (%) info Average percentage of 100 test cases successfully resolved across 10 runs for each model
arrow_range Cl range (%) info Expected performance range, reflecting the results' statistical reliability (p-value < 0.05)
Avg latency (h) info Average time taken to solve 100 tasks across 10 runs
Avg total tokens (M) info Average token consumption for a full benchmark run (100 tasks) across 10 runs
Avg cost ($) info Average cost per full benchmark run
Date
GPT 5.5 74.0
66.8 — 80.5 15.5 64.5 $133.9 2026-04-27
GPT 5.4 72.4
65.4 — 79.3 21.2 64.2 $91.7 2026-03-16
Gemini 3.1 Pro Preview 72.4
65.1 — 78.8 11.5 75.4 $49.0 2026-02-27
Claude Opus 4 7 68.7
60.5 — 75.9 11.6 90.0 $124.3 2026-04-27
GPT 5.3 Codex 67.7
59.9 — 75.6 11.2 71.4 $42.6 2026-03-18
Claude Opus 4 6 66.6
59.1 — 74.1 9.9 69.5 $84.4 2026-02-26
GPT 5.2 Codex 62.5
54.4 — 70.0 24.3 124.4 $121.9 2026-02-27
Claude Opus 4.5 61.9
53.9 — 70.2 12.5 79.8 $102.5 2026-02-26
Gemini 3 Pro Preview 60.4
52.3 — 67.7 9.8 117.0 $63.7 2026-02-27
GLM 5.1 59.7
52.4 — 67.4 33.4 80.2 $46.7 2026-05-08
Claude Sonnet 4.6 58.4
50.3 — 66.4 8.2 47.9 $40.4 2026-03-01
Kimi K2.6 58.6
51.3 — 66.5 29.9 94.3 $42.5 2026-05-10
DeepSeek V4 Pro 55.4
47.5 — 63.6 35.8 132.7 $13.7 2026-05-08
Claude Sonnet 4.5 54.2
45.9 — 62.2 13.1 92.9 $60.3 2026-02-26
DeepSeek V4 Flash 52.7
45.3 — 60.7 28.1 164.7 $8.4 2026-05-11
MiMo 2.5 Pro 52.0
43.8 — 60.0 33.1 97.5 $74.5 2026-05-09
Qwen 3.6 Max Preview 51.4
43.5 — 59.3 20.5 103.0 $222.4 2026-05-07
Gemini 3 Flash Preview 42.0
36.6 — 47.3 16.5 148.0 $34.2 2026-02-26
MiniMax M2.7 37.2
30.3 — 44.9 20.3 128.3 $10.1 2026-05-01
Qwen 3.6 27B 37.4
30.5 — 44.5 20.7 112.3 $64.6 2026-05-05
Gemma 4 31B IT 33.2
26.2 — 40.8 14.2 29.5 $2.5 2026-05-01
Qwen 3.6 35B A3B 31.7
24.4 — 39.0 12.5 113.4 $10.7 2026-05-05
Gemini 2.5 Pro 29.1
22.3 — 36.1 8.4 37.9 $35.8 2026-03-02
Gemma 4 26B A4B IT 25.1
18.8 — 31.8 21.4 77.2 $3.3 2026-05-01
GPT OSS 120B 18.9
13.1 — 25.1 25.9 122.7 $7.6 2026-05-09
Gemini 2.5 Flash 15.9
10.7 — 21.1 4.9 108.8 $11.2 2026-02-26
Qwen 3.5 9B 15.5
10.1 — 20.9 16.6 181.4 $15.6 2026-05-07
GPT OSS 20B 2.4
1.2 — 3.9 3.8 12.0 $0.2 2026-05-11
Latest results as of May 18th 2026: This refresh includes open-weight models, adding new columns for latency, tokens, and cost.
Check back periodically for updates!

Learn more about Android Bench

Learn more about how we created a set of common Android developer tasks.
Many of the tasks are based on how we define high quality Android development, which is detailed in our developer documentation.
See the full repo so you can replicate the tests yourself.