OpenAI is taking its most capable model to court, in the professional sense. Astra for Law, introduced Thursday, is a legal configuration of GPT-6 Astra.
The centerpiece is a search index covering more than 230 million URLs of U.S. legal material. Around it sit access controls and workflow tools for law firms and legal technology companies. API customers Harvey and Legora are named as builders on the platform.
Deployment starts narrow. Selected firms get access through Trusted Access in ChatGPT and Codex, and a separate API release follows later.
Scope covers case law, statutes, regulations, court rules and administrative decisions, refreshed daily. A partnership with the Free Law Project, the nonprofit behind CourtListener, contributes a case-law collection spanning more than 99.9 percent of published U.S. precedential case law. OpenAI positions the index as complementary to licensed material firms already buy, naming Thomson Reuters.
The yardstick behind the launch numbers is Vals AI’s Legal Research Bench. OpenAI drew 200 questions from its private validation set, which grades source and passage retrieval along with answer quality.
The legal configuration passed the overall correctness check on 54.0 percent of questions. GPT-6 Astra using web search alone passed 38.7 percent, which OpenAI calls a 40 percent relative improvement.
Two further gaps were reported: 24 percent more reference cases, and up to 54 percent more relevant passages from the right opinions.
The base model already clears the Critical cybersecurity threshold in OpenAI’s Preparedness Framework, and reaches ChatGPT paid tiers, the API, Microsoft Azure and AWS Bedrock. Legal is the first vertical carved out of it.
