OpenAI introduced Astra for Law on September 17, 2026, as a legal-work configuration of GPT-6 Astra. It combines the model with legal-analysis and writing instructions, configurable settings, tools and a dedicated Legal Search Index. Initial access is limited to selected U.S. law firms through Trusted Access in ChatGPT and Codex, while API access is coming soon.
The headline benchmark is an OpenAI-reported 54.0% overall correctness pass rate on 200 U.S. legal research questions at the highest reasoning effort. GPT-6 Astra using web search alone scored 38.7% under the same stated conditions. Those figures describe a specific U.S. legal-research evaluation, not every legal task or jurisdiction.
OpenAI’s Astra for Law is a legal configuration of GPT-6 Astra
Astra for Law is not presented as a separate base model. OpenAI describes it as GPT-6 Astra configured for professional legal work, with instructions and settings designed for legal analysis and writing.
The product’s architecture brings several pieces together: the underlying model, a Legal Search Index, configurable tools and legal-work guidance. The intended workflow is to move from the facts of a matter to relevant authorities and supporting passages that lawyers can examine as part of their work.
In ChatGPT, OpenAI says the model appears as GPT-6 Astra Law. Its API identifier is gpt-6-astra-law.
What the Legal Search Index covers
OpenAI says the Legal Search Index searches more than 230 million URLs across U.S. case law, statutes, regulations, court rules and administrative decisions, with sources added daily.
OpenAI also says the CourtListener collection, developed with Free Law Project, brings coverage of more than 99.9% of published U.S. precedential case law into the research experience. That description applies specifically to published precedential case law; it is not a blanket figure for every kind of U.S. legal material.
The distinction matters in practice. A large search corpus can improve the path to relevant authority, but a lawyer still has to examine the cited material and its fit for the matter at hand.
What the 54% benchmark result actually measures
OpenAI evaluated Astra for Law on 200 U.S. legal research questions from Vals AI’s private Legal Research Bench validation set. At the highest reasoning effort, Astra for Law passed the overall correctness check on 54.0% of the questions. GPT-6 Astra with web search alone passed on 38.7%.
| System | Overall correctness pass rate | Evaluation set and condition |
| Astra for Law | 54.0% | 200 U.S. legal research questions at the highest reasoning effort |
| GPT-6 Astra with web search alone | 38.7% | The same 200-question set and reasoning condition |
OpenAI describes the difference as a 40% relative improvement. The calculation from the reported rates is approximately 39.5%, which rounds to 40%.
The company also reports 24% more reference cases on case-law-focused questions and up to 54% more relevant passages from correct court opinions on an audited target-passage set. These are scoped measurements attached to particular evaluation tasks. They do not turn the system into a universal legal-accuracy guarantee.
Where firms and legal-tech companies fit
Astra for Law is aimed at more than litigation research. OpenAI documents transactional workflows built around firm playbooks and selected precedents.
Sullivan & Cromwell built an agreement analyzer that identifies risks created by provisions considered together, proposes redlines and drafts client advice for lawyer review. Ropes & Gray built a transaction-diligence system that reviews data-room materials, traces findings to source documents and flags questions such as whether customer contracts require notice or consent.
Cooley built GO Public for IPO preparation and review. The workflow incorporates capital-markets expertise into filing analysis and carries deal changes across the filing so lawyers can inspect their implications together.
The ecosystem also includes 26 partner-built plugins and 47 community skills for legal work in ChatGPT. OpenAI identifies Harvey and Legora as API customers that will be able to build on Astra for Law. Integrations or plugins involving Relativity, Clio, iManage, Intapp, DeepJudge and Thomson Reuters are also part of the launch ecosystem.
Access, privacy and the limits of the launch
Astra for Law is initially available to selected U.S. law firms through Trusted Access in ChatGPT and Codex. That means it is not an open product rollout for every U.S. law firm or every ChatGPT user. OpenAI describes API access as coming soon.
For eligible legal API use, OpenAI says Zero Data Retention applies. Under its legal-industry offering, ChatGPT Enterprise usage is excluded from human review by default.
Those controls address how firms can manage sensitive legal work, but they do not remove the professional task of checking an answer. OpenAI’s help documentation tells users to review answers and cited sources before relying on them. For a legal workflow, that review is the point at which a generated research path becomes work a lawyer can assess and use.
