Google releases updated android AI benchmarks featuring kimi K3.1 pro and DeepSeek V4.1 flash scores

Google officially published updated benchmark scores for three AI models—OpenAI GPT-6.1 Terra, Kimi K3.1 Pro, and DeepSeek V4.1 Flash—on its Android Developers channel on November 10, 2026. The Android Bench is a specialized AI programming benchmark developed by Google, designed to assess models' capabilities in real-world engineering tasks such as application creation, feature development, code migration, and application conversion. Kimi K3.1 Pro, evaluated using the kimi-code agent, achieved a 24.0% pass rate across 30 long-form tasks, with an average completion rate of 82.3% and an average runtime of 27.4 hours, costing $317.4 per run. While Kimi K3.1 Pro outperformed GPT 6 Astra, Claude Fable 5.1, and Gemini 4 Argon, its score fell short of GPT 6.1 Terra's 28.7%. DeepSeek V4.1 Flash, tested with the opencode agent, recorded a 17.3% pass rate and 71.2% average completion, taking 12.0 hours and incurring a cost of just $9.9 per benchmark run. This model offers significant advantages in speed and cost efficiency, making it suitable for budget-conscious development teams.