en.Wedoany.com Reported - Google has updated the Android Bench benchmark, which is used to evaluate the performance of large language models in Android development tasks. Google engineers have redesigned the testing methodology, added new models, and allowed developers to contribute to the development of the test set.

The main update involves switching to the Harbor framework, which now executes all Android Bench tests. This change ensures more consistent model evaluations, reproducible results, and faster addition of new test scenarios. Additionally, the new system not only evaluates answer quality but also assesses the cost and efficiency of specific LLMs.

Third-party developers can now influence the direction of the benchmark. Google encourages developers to share real-world tasks and discuss model evaluation methods, which helps regularly update the test set and more accurately reflect LLM usage scenarios in Android development.

Several new language models have been added to the results table, including Claude Fable 5, Claude Sonnet 5, Claude Opus 4.8, GLM 5.2, Kimi K2.7 Code, MiniMax M3, Qwen 3.7 Max, and Qwen 3.7 Plus. The current top three are Claude Fable 5, GPT 5.5, and Claude Sonnet 5.