Android Bench 2.0 Completion
Code · 2026-09-17
Mean partial completion across the same 30 Android Bench 2.0 tasks and five trials. Measures progress toward satisfying all task checks, including failed tasks; shares its factor group with full-task pass rate.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Opus 5.5 | 84.7 |
| Claude Fable 5.1 | 82.4 |
| GPT-6 Astra | 82.2 |
| Kimi K3 | 72.8 |
| GPT-6 Sol | 70.2 |
| Sonnet 5 | 59.8 |
| GPT-6 Luna | 53.2 |
| Gemini 3.7 Flash | 50.1 |
| Gemini 3.8 Flash | 47.4 |