大模型
www.aibase.com ·
OpenAI 推出 Astra for Law：索引 2.3 亿 URL，法律问答准确率 54%
OpenAI 发布面向法律领域的 AI 基础平台 Astra for Law，法律检索索引覆盖超 2.3 亿 URL，并纳入覆盖超 99.9% 已公开美国判例的 CourtListener 数据库。在 Vals AI 法律研究基准的 200 道私有测试题上，最高推理强度下整体准确率为 54.0%，仅依赖网页搜索的 GPT-6Astra 为 38.7%，差距超 15 个百分点；判例类问题多找到 24% 相关案例，在经审核的目标段落集中从正确判决书中多提取最多 54% 相关段落。OpenAI 强调其并非“AI 律师”，主要服务律师与法律软件公司，用于区分判决核心推理与附带意见、主动查找不利先例、分析合同条款风险分配。
AI推荐理由 OpenAI 发布面向法律领域的 Astra for Law，正文给出具体数据：法律检索索引覆盖超 2.3 亿 URL，含 CourtListener 判例库，覆盖超 99.9% 已公开美国判例；在 Vals AI 法律研究基准的 200 道私有测试题上最高推理强度准确率 54.0%，仅靠网页搜索的 GPT-6Astra 为 38.7%，差距超 15 个百分点，判例类问题多找到 24% 相关案例。这些是具体可核验的指标，对法律 AI 落地有参考价值。局限：来源为 AIbase 转述 OpenAI 博客，非一手原文；54% 绝对准确率仍偏低，且测试集为私有、规模仅 200 题，不能据此推断实际办案效果；OpenAI 明确其不是“AI 律师”，主要服务律师与法律软件公司。
72
正文
OpenAI Launches Astra for Law: Indexes 230 Million URLs, Legal Q&A Accuracy of 54%, Outperforms General Models by 15 Percentage Points
AIbase基地
Published inAI News · 3 min read · Sep 18, 2026
8
On September 17, OpenAI published a blog post introducing Astra for Law, an AI foundation platform designed specifically for legal use. Its confidence comes from the scale of its data: the legal search index covers more than 230 million URLs and includes the CourtListener case database, which covers over 99.9% of published U.S. case law—effectively bringing almost all publicly available U.S. case law into the model's retrieval scope.
The performance was validated using a private test set from the Vals AI legal research benchmark, consisting of 200 U.S. legal questions. At the highest reasoning intensity, Astra for Law achieved an overall accuracy rate of 54.0%, while GPT-6Astra, which relied only on web searches, reached 38.7%, a difference of more than 15 percentage points. Specifically, in case law-oriented questions, Astra for Law found 24% more relevant cases; in audited target paragraph sets, it extracted up to 54% more relevant paragraphs from correct court opinions.
However, OpenAI emphasized that this is not an "AI lawyer." It primarily serves lawyers and legal software companies, performing several specific tasks: helping distinguish core reasoning from incidental comments in court decisions, actively finding unfavorable precedents, and analyzing how contract clauses allocate risks—focusing on increasing the chances of winning a case in court. In other words, it won't appear in court for you, but it will show you every possible gap in the precedents.
AstraforLawOpenAIAI Legal PlatformCourtListener
This article is from AIbase Daily
Scan to view
Welcome to the [AI Daily] column! This is your daily guide to exploring the world of artificial intelligence. Every day, we present you with hot topics in the AI field, focusing on developers, helping you understand technical trends, and learning about innovative AI product applications.
—— Created by the AIbase Daily Team
© Copyright AIbase Base 2024, Click to View Source -