OpenAI Launches 'Astra for Law', a Legal-Specialized AI Based on GPT-6

On September 17, 2026, OpenAI unveiled 'Astra for Law', a legal-specialized foundation built on GPT-6 Astra. This model combines a robust legal search index with analytical instructions, designed for law firms and legal tech companies to integrate into their own workflows.

Key Performance and Technical Composition

Astra for Law features a built-in legal search index capable of retrieving US case law, statutes, and regulations. Through a partnership with the Free Law Project, it covers over 99.9% of published US precedents. Validation results from the Vals AI Legal Research Bench show an overall accuracy pass rate of 54.0%, representing a 40% improvement over GPT-6 Astra using web search only (38.7%). Additionally, it discovered 24% more relevant precedents in case-centric queries.

Partner Integration and Governance

API customers such as Harvey and Legora, along with major law firms including Sullivan & Cromwell and Ropes & Gray, participated in early testing. Sullivan & Cromwell developed negotiation playbooks and an agreement analyzer, while Ropes & Gray developed a deal diligence system. Through the Trusted Access Program, it offers privacy and governance features such as confidential work controls for law firms, Zero Data Retention (ZDR), and default exclusion of human review.

Ecosystem Expansion and Roadmap

It connects with existing professional tools through 26 new ecosystem plugins, including Relativity and Clio. Nine community plugins and 47 custom skills have been released, broadening the range of tools available to legal professionals. OpenAI plans to maintain an open and composable architecture, aiming to improve the model long-term so that law firms can apply their own expertise and standards to the AI.