OpenAI's Astra for Law jumps GPT-6 legal accuracy from 39% to 54%

OpenAI has launched Astra for Law, a version of its top model, GPT-6 Astra, set up for legal work. It comes with legal instructions and a dedicated legal search index. The index covers more than 230 million URLs of US case law, statutes, regulations, court rules and administrative decisions, and new sources are added daily. The case law comes from Free Law Project, the nonprofit behind CourtListener, whose collection covers more than 99.9% of published US precedential case law.

The gains are real but modest. On Vals AI's Legal Research Bench (200 questions), Astra for Law passed the correctness check on 54% of questions. Plain GPT-6 Astra with web search passed 38.7%, so that's a 40% relative jump. It also found 24% more reference cases and up to 54% more relevant passages from the right court opinions. Still, it failed nearly half the questions, so every answer needs checking.

Access is limited for now. Lawyers at eligible firms, mainly the Am Law 200, can get in through a new Trusted Access Program. API customers like Harvey and Legora will be able to build on the model "soon", but OpenAI hasn't given a date for wider release.

Why it matters: OpenAI is moving from supplying models to legal-tech startups to selling legal AI to big firms itself.