AI 导读
Artificial Analysis 搜索指数如何运作：候选模型运行在 Stirrup 中——这是我们开源的智能体框架，配备从网上搜索和抓取页面的工具。模型每个任务最多有 25 轮。当它获得足够信息后，会提交最终答案，由特定基准的评分器为答案打分。 每个提供商都运行完全相同的智能体和基准设置，唯一的变化是网页搜索工具背后是哪家提供商。
正文
How the Artificial Analysis Search Index works: the candidate model runs inside Stirrup, our open-source agent harness, with tools for searching and fetching pages from the web. The model has up to 25 turns per task. When it has enough information, it submits its final answer and a benchmark-specific grader scores the answer.
Every provider runs with the exact same agent and benchmark setup, with the only change being which provider sits behind the web search tool.
来源：@ArtificialAnlys · x.com