Google updated Android Bench today, adding eight new AI models to its Android app development benchmark leaderboard. The update includes Claude Fable 5, Claude Sonnet 5, Claude Opus 4.8, GLM 5.2, Kimi K2.7 Code, MiniMax M3, Qwen 3.7 Plus, and Qwen 3.7 Max. Google created the benchmark to evaluate how large language models perform across 100 Android development tasks, with the original version launching in March.
Google Adds Eight AI Models to Android Bench Leaderboard
Google expanded Android Bench with eight new models. The additions include Claude Fable 5, Claude Sonnet 5, Claude Opus 4.8, GLM 5.2, Kimi K2.7 Code, MiniMax M3, Qwen 3.7 Plus, and Qwen 3.7 Max. The company also adopted a new framework and added metrics for cost and efficiency. Google invited developers to run their own tests and submit feedback.
Claude Fable 5 Leads Benchmark at 84.5 Percent Accuracy
Claude Fable 5 achieved 84.5 percent accuracy in the Android Bench test, placing first on the leaderboard. GPT 5.4 and Claude Sonnet 5 ranked second and third. Gemini 3.1 Pro placed fifth in the updated rankings. OpenAI's models held a slight lead over Google's AI in the initial release launched in March.
Android Bench Evaluates 100 Development Tasks
Android Bench tests large language models across a suite of 100 Android development tasks. The benchmark aims to demonstrate which AI agents perform best at Android app development. Google launched the initial version in March and has since added open-weight models to the evaluation framework.
FAQ
What models did Google add to Android Bench today?
Google added eight new models to Android Bench: Claude Fable 5, Claude Sonnet 5, Claude Opus 4.8, GLM 5.2, Kimi K2.7 Code, MiniMax M3, Qwen 3.7 Plus, and Qwen 3.7 Max.
What accuracy did Claude Fable 5 achieve in Android Bench?
Claude Fable 5 achieved 84.5 percent accuracy in the Android Bench test, placing first on the leaderboard.
When did Google launch the original Android Bench?
Google launched the initial version of Android Bench in March.