Google updates android bench rankings: gemini 4 argon lags behind claude and GPT in programming capabilities

On October 28, 2026, Google updated its Android Bench rankings, which evaluate AI models' capabilities in Android programming assistance. The latest results show that Gemini 4 Argon significantly lags behind the latest versions of Claude and GPT models. The Android Bench was launched earlier this year to help Android developers choose the best AI models for code development. The test considers Android-specific use cases and best practices, and Google has included average call cost as a benchmark metric to aid developers in their selection. In the October 26 update, Google added a new label for each model's inference level and included additional models such as GPT-6.1 Sol and Claude Sonnet 5.5. The top performer is Claude Sonnet 5.5 with a 40% pass rate and 80.9% completion rate. GPT-6.1 Sol, a new addition, has a 29.3% pass rate and 82.8% completion rate, with a call cost of $99.50. In contrast, Claude Sonnet 5.5 has a much higher call cost of $1,574.90. Gemini 4 Argon, Google's own model, ranked fifth with a 21.3% pass rate and 81.8% completion rate, and a call cost of $267.70. Google acknowledged the performance of Gemini 4 Argon despite internal reports suggesting it struggles with some programming tasks.