OpenAI launches Astra for Law, a legal-focused GPT-6 Astra model

OpenAI launched Astra for Law on 17 September 2026, a version of its GPT-6 Astra model configured with settings, tools and context for professional legal work.

OpenAI said the tool combines its most powerful model with a Legal Search Index covering more than 230 million URLs of US case law, statutes, regulations, court rules and administrative decisions, updated daily. The product appears as "GPT-6 Astra Law" in the model picker.

On a 200-question test drawn from the Vals Legal Research Bench, OpenAI said Astra for Law scored 54.0% correct, against 38.7% for GPT-6 Astra using web search. The company said it also found 24% more relevant cases on case-law questions. OpenAI has not published an independent comparison against incumbent legal research platforms.

Case law in the index comes from a partnership with the nonprofit Free Law Project, via its CourtListener database, which OpenAI said covers more than 99.9% of published US precedential case law.

The launch includes 26 partner plugins from vendors including Thomson Reuters, Intapp, Relativity, Clio and DeepJudge, alongside nine community plugins offering 47 skills. ChatGPT for Word is now generally available, according to OpenAI.

Eligible firms will get zero data retention and default exclusion from human review, OpenAI said. Access is initially limited to selected US law firms through OpenAI's Trusted Access and Codex programmes, with an API version coming soon.

Early adopters named by OpenAI include Latham & Watkins, Ropes & Gray, Cooley and Sullivan & Cromwell. Legal technology firms Harvey and Legora are building Astra for Law into their own products, according to OpenAI. Ropes & Gray has used the tool in an M&A due-diligence system that traces findings to source documents and checks whether customer contracts require notice or consent for an acquisition, Artificial Lawyer reported.

OpenAI said an API version of Astra for Law is coming soon, which would extend access beyond the firms currently in its Trusted Access and Codex programmes.