Event ·Multi-source

谷歌推出 Android 开发模型基准测试 Android Bench 2.0

谷歌推出 Android Bench 2.0 基准测试，专门衡量模型完成真实 Android 开发工作的能力。测试对象是“模型+Agent”组合，如 GPT + Codex、Claude + Claude Code 等。测试包含 30 个真实工程任务，覆盖应用转换、从零构建、架构迁移和新功能开发四类。评测结果显示，GPT 6 Astra + Codex 和 Claude Fable 5.1 + Claude Code 分别位列前两名，Google 自家的 Gemini 3.8 Flash 和 3.7 Flash + Antigravity 排名靠后。部分任务如 Flutter/React Native 转换和 XML→Compose 迁移，所有模型均未通过。

Current read

谷歌推出的这个新基准测试挺有意思，专门测模型做真实 Android 开发的能力，比如从零写应用或者迁移代码，能看出不同模型在工程场景下的实际表现。

11 sources78° AI heatLast updated 2026/9/19 13:09:36

Evidence

11 sources

Primary sourceshao__meng

2026/9/19 13:09:36View source

Related source 1Decoder

2026/9/19 13:28:55View source

Related source 2IT之家

2026/9/20 05:46:36View source

Related source 3歸藏(guizang.ai)

2026/9/21 03:06:58View source

Related source 4AI Will

2026/9/20 02:43:28View source

Related source 5Gary Marcus

2026/9/17 14:46:38View source

Related source 6The Rundown AI

2026/9/17 19:02:20View source

Related source 7OpenAI

2026/9/17 20:15:15View source

Related source 8a16z

2026/9/17 21:19:09View source

Related source 9@OpenAIDevs

2026/9/19 00:56:31View source

Related source 10Jerry Liu

2026/9/19 01:07:43View source

Conflict check

Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.

Open the story