Source: explainx.ai

OpenAI introduced Astra for Law on September 17, 2026, deploying GPT-6 Astra configured for legal practice by adding a 230-million-source U.S. legal search index, instructions tuned to legal analysis and writing, and firm-specific customization built with forward-deployed engineers. The Legal Search Index aggregates U.S. case law, statutes, regulations, court rules, and administrative decisions with daily updates and incorporates CourtListener via a Free Law Project partnership that OpenAI says covers more than 99.9% of published U.S. precedential case law. OpenAI reports performance on 200 U.S. legal-research questions from Vals AI’s Legal Research Bench: Astra for Law passed the bench’s overall correctness check on 54.0% of questions versus 38.7% for web-search GPT-6 Astra, a relative improvement OpenAI states as 40%.

Explainx.ai highlights supporting metrics (24% more reference cases on case-law questions; up to 54% more relevant passages on an audited subset), the launch ecosystem (26 partner-built plugins, 47 community plugins, and named collaborators including Thomson Reuters, Sullivan & Cromwell, Ropes & Gray, Cooley, Harvey, and Legora), and Trusted Access privacy features (Zero Data Retention on the API and ChatGPT Enterprise usage excluded from human review by default). The piece cautions that the benchmark results were OpenAI’s own test on a private validation set with no disclosed independent reproduction at launch, and that the two worked examples comparing Astra for Law to Claude Fable 5.1 are hand-picked marketing examples without published selection methodology. Explainx.ai frames Astra for Law as a vertical product template—frontier general model + current domain retrieval + domain instructions + enterprise access controls—while noting selective Trusted Access, dependence on the underlying GPT-6 Astra model, and the absence of independent audit as important limitations for firms to weigh.