DATA CUT 2026-07-27 116 活跃模型 680 A 类案例

CASE EVIDENCE / A RECORD

Cognition 使用 o1-preview 处理软件工程任务执行

Cognition 公开的代码代理与软件工程案例,来源为 博客记录,复核于 2026-06-27T09:42:50Z。

原始记录:Cognition tested o1-preview inside Devin for root-cause debugging and autonomous coding-agent tasks

A

Chinese Brief

中文案例导读

Cognition 公开的代码代理与软件工程案例,来源为 博客记录,复核于 2026-06-27T09:42:50Z。 Model Atlas 将它标记为 A 类证据,因为它同时具备具体使用者、具体任务、公开原始证据和可访问产物。 Model Atlas 不把 benchmark、教程、发布说明或集合页包装成真实案例。

厂商OpenAI
模型o1-preview
任务类型代码代理与软件工程
审核状态auto_approved

任务

真实任务背景

这是一个围绕软件工程任务执行的真实任务,公开材料可以回溯到具体使用者和具体产物。 原始资料写作:Cognition evaluated OpenAI o1-preview in Devin, its autonomous software engineering agent, including tasks that require installing libraries, browsing for context, editing and running code, diagnosing dependency errors,…

代码代理与软件工程博客记录A 类可核验real_case
公开产物

公开材料提供原始证据链接和可访问产物,可用于核验任务结果、项目形态和模型绑定关系。 原始资料写作:Cognition reported that replacing key Devin-Base subsystems with the o1 series produced significant gains on its internal cognition-golden evaluation suite; in a concrete sentiment-analysis task, Devin with o1-preview c…

模型作用

o1-preview 在该案例中承担软件工程任务执行相关的生成、分析、编排或实现角色。 原始资料写作:Cognition attributes the improvement to o1-preview's ability to reflect, analyze, backtrack, diagnose root causes instead of symptoms, and research online like a human engineer when resolving complex upstream causes in …

风险边界

当前判断基于公开材料;若产物下线、仓库变更或模型参与比例仅来自作者自述,需要在引用前重新复核。 原始资料写作:The blog is an evaluation and product-engineering writeup rather than a third-party customer deployment; it explicitly names o1-preview and includes concrete Devin task artifacts, but some aggregate benchmark gains are …