OpenAI Launches Astra for Law, With Harvey and Thomson Reuters as Launch Partners

OpenAI launched Astra for Law today, a version of its GPT-6 Astra model built specifically for legal research and writing. Latham & Watkins, Ropes & Gray, Cooley, and Sullivan & Cromwell are already using it.

The product centers on a new legal search index covering U.S. case law, statutes, regulations, court rules, and administrative decisions across more than 230 million URLs, with sources added daily. OpenAI built part of that index with Free Law Project, the nonprofit behind CourtListener, whose case-law collection covers more than 99.9% of published U.S. precedential case law.

The numbers OpenAI is citing

OpenAI tested the tool against 200 legal research questions from Vals AI's Legal Research Bench, a benchmark that scores how well a model finds relevant sources and meets research-quality criteria.

At the highest reasoning setting, Astra for Law passed the benchmark's correctness check on 54% of questions, compared with 38.7% for the same underlying model using ordinary web search — a 40% relative improvement. On case-law-specific questions, it surfaced 24% more reference cases and pulled up to 54% more relevant passages from the correct opinions.

What firms are actually building with it

Sullivan & Cromwell built an agreement analyzer that checks new deals against the firm's own negotiating playbooks and past precedents, then drafts proposed redlines. Ropes & Gray built a data-room review tool that traces each finding back to its source document and flags issues like missing contract consents. Cooley built a tool called GO Public that helps draft IPO filings and updates the filing automatically when deal terms change.

The competitors are plugin partners, not casualties

The part worth paying attention to if you're evaluating legal AI tools right now: OpenAI didn't try to route around the existing legal AI vendors. Astra for Law launches with 26 partner-built plugins and 47 community plugins, and the partner list includes Thomson Reuters, Harvey, Legora, and iManage — companies that could reasonably be read as OpenAI's competitors in this space.

Thomson Reuters' own CTO framed it as complementary rather than competitive, positioning CoCounsel Legal as the layer that adds enterprise context and governance on top of what a general model provides. Harvey's Head of Applied Research praised Astra for Law's grounding and citation precision in the same announcement where it appears as a listed partner.

OpenAI is also running a Trusted Access Program for eligible firms, offering zero data retention on API usage and excluding ChatGPT Enterprise activity from human review by default — working directly with Latham & Watkins on permissions, ethical walls, and firm-level oversight controls.

What this means for tool buyers

The near-term read isn't that a general AI lab is about to displace specialist legal research vendors. It's that the specialist vendors are moving to plug into the general model rather than compete head-on with it, which suggests the market is settling into layers — a foundation model handling raw research, and specialist products adding the firm-specific context, governance, and workflow on top. For anyone choosing legal AI tools this year, the question worth asking a vendor is less "does this replace ChatGPT" and more "how does this connect to whatever model layer eventually wins."

Sources:

• Artificial Lawyer — "OpenAI Launches Astra For Law"
• Artificial Lawyer — "Astra for Law's 26 Legal Tech Plugins"