Google launches Android Bench 2.0 testing AI on Android development
Technology
Google has rolled out Android Bench 2.0, a tool that puts AI models through tough, real-world Android development challenges, like updating dependencies or building entire apps from scratch.
The new version ditches simple pass-fail scoring for a more detailed system, so you get a clearer picture of what these AIs can actually do.
GPT-6 Astra leads Android bench 28%
GPT-6 Astra came out on top with a 28% pass rate, while Gemini 3.8 Flash managed just 8%.
If you're curious which AI is best for your next project, Google's updated leaderboard will keep tracking new results as more models are added.