Gemini 4 Argon trails Claude and GPT in Google's Android coding test
Android Bench is Google's leaderboard that tests how well AI models assist with Android coding. It looks at Android-specific tasks and best practices, and now factors in average API costs. The list was recently updated to show each model's reasoning level.
Claude Sonnet 5.5 now leads with a 40% pass rate and 80.9% completion rate. OpenAI's GPT-6.1 Sol scored a 29.3% pass rate and 82.8% completion rate, costing only $99.50, far less than Claude's $1,574.90.
Google's own Gemini 4 Argon had a weak debut, ranking fifth with just a 21.3% pass rate, well behind other top models. Leaks earlier suggested it struggles with some coding tasks, but Google still backs it.
Conversational AI that understands complex instructions, generates long-form answers and plugs into any workflow
Chat freely with an advanced AI that drafts emails, debugs code, brainstorms ideas and simplifies everyday tasks
Your personal, proactive Google AI assistant that simplifies work, school and everyday life right on your phone
ByteDance AI chatbot development platform with LLM integration, plugin ecosystem and cross-platform deployment to Doubao and Feishu
Comments
Loading comments…