Google revamps Android AI dev benchmark, adds Fable 5 and other agents

Google has introduced significant updates to its Android Bench, a benchmark designed to evaluate large language models (LLMs) specifically for app development tasks. The revamped benchmark now features a total of eight new models, expanding the competitive landscape for AI in software creation. OpenAI's GPT 5.4 and Anthropic's Claude Fable 5 have emerged as the top performers in terms of accuracy, according to the latest leaderboard. This represents a shift in the rankings, with Google's Gemini 3.1 Pro now positioned in fifth place. Beyond just accuracy, the updated Android Bench also incorporates crucial factors such as cost and efficiency into its evaluation criteria, reflecting a more holistic approach to assessing LLM performance in real-world development scenarios.