OpenAI has introduced Astra for Law, a version of GPT-6 Astra connected to a specialized index of US case law, statutes and regulations. The product combines that search layer with instructions tailored to legal analysis and writing.
The index spans more than 230 million URLs and uses material from the Free Law Project, which says its collection includes more than 99.9 percent of published US precedents. In an OpenAI-run evaluation using the 200-question Legal Research Bench from Vals AI, Astra for Law answered 54 percent successfully. Standard GPT-6 Astra with ordinary web search scored 38.7 percent. Because OpenAI conducted the comparison, law firms will still need independent tests on their own jurisdictions, citation requirements and workflows.
API customers including Harvey and Legora can build services on the model. OpenAI is also releasing 26 integrations for legal tools such as Relativity and Clio, plus a Trusted Access program for law firms with privacy controls including zero data retention. Better source retrieval can reduce one cause of legal hallucinations, but a 54 percent benchmark result leaves substantial room for error. Lawyers remain responsible for checking authorities, quotations, jurisdiction and whether a model’s reasoning fits the facts of a client’s matter.