OpenAI Launches Astra for Law, Pairing GPT-6 Astra With a Legal Search Index

OpenAI on September 17 introduced Astra for Law, a version of its GPT-6 Astra model paired with a large legal search index and legal-analysis instructions, offered first to selected law firms, SiliconANGLE reported. It is not a new model. That is the point for enterprise leaders: the gains OpenAI is claiming come from retrieval, domain data, connectors and governance controls built around a general model, the same layers that decide whether AI works in any regulated function.

What is Astra for Law?

OpenAI describes Astra for Law as a configuration of GPT-6 Astra, which began rolling out on September 3. Its search index covers more than 230 million URLs spanning U.S. case law, statutes, regulations, court rules and administrative decisions, with sources added daily, according to Unite.AI. Case law comes through a collaboration with Free Law Project, the nonprofit behind the CourtListener database, which OpenAI says covers more than 99.9% of published U.S. precedential decisions. OpenAI says the index complements licensed providers such as Thomson Reuters rather than replacing them.

Access is limited at first. Selected firms can use it through a Trusted Access program in ChatGPT and Codex, where it appears as "GPT-6 Astra Law." An API version is planned with no announced date or pricing, and legal software companies Harvey and Legora are named as customers that will build on it.

Eligible firms get zero data retention on the API, and ChatGPT Enterprise usage is excluded from human review by default, according to SiliconANGLE.

How well does Astra for Law perform?

OpenAI tested 200 questions from the private validation set of Vals AI's Legal Research Bench. Astra for Law passed the overall correctness check on 54.0% of questions, compared with 38.7% for GPT-6 Astra using web search alone, with both at maximum reasoning effort. OpenAI also reported finding 24% more reference cases on case-law questions and up to 54% more relevant passages on a separate audited set.

Those figures need context. They are OpenAI's own results, the sample is small, and the baseline is OpenAI's own model with web search, not a competing legal research product, as Unite.AI noted. A 54% pass rate also means the system failed the correctness check on nearly half the questions.

Firms should test on their own matters. "Most of the time your data isn't public," HumanSignal ML Evangelist Micaela Kaplan said at COLLIDE 2025, explaining why custom benchmarks beat leaderboards. Public questions cannot show how a model handles a firm's own playbooks.

Which law firms and vendors are involved?

Latham & Watkins is working with OpenAI on governance design, including information permissions, ethical walls and client instructions. OpenAI engineers also built firm-specific tools on ChatGPT Enterprise: an agreement analyzer at Sullivan & Cromwell, a deal diligence system at Ropes & Gray and GO Public, an IPO preparation tool, at Cooley. Sullivan & Cromwell partner John Savva said early versions showed "impressive research depth and sensitivity to authority," according to Unite.AI.

The launch came with 26 partner-built plugins connecting ChatGPT to tools such as Relativity, Clio, iManage, Intapp and DeepJudge, plus nine community plugins adding 47 custom skills. ChatGPT for Word became generally available the same day.

Incumbents are positioning as partners. Thomson Reuters is bringing context from its HighQ platform into ChatGPT and previewing a connector for CoCounsel Legal. Its CTO, Joel Hron, said "the value is not in connectivity alone," stressing trusted intelligence and governance, according to SiliconANGLE.

Are AI labs moving into vertical markets?

Astra for Law was one of two industry launches from frontier labs this week. On September 14, Anthropic released Claude for Financial Advisors, a package of connectors and workflow skills linking Claude to custodians, portfolio platforms, CRM systems and wealth planning tools, Silicon Report reported. The skills cover meeting preparation, portfolio analysis, follow-up drafts, pipeline review and compliance checks, including a skill that screens client-facing language and tracks AI usage under SEC rules. Advisers must review and approve critical tasks. The same report noted that OpenAI launched ChatGPT for Financial Services on September 10.

Labs are packaging industry data, connectors and compliance workflows on top of general models. That pressures specialized vendors, which build on the same models they compete with, and shifts the differentiator toward each customer's own knowledge. "The context layer for your organization is where all your alpha is going to accrue," Patrick Gallic, Head of International AI Hub at Tokio Marine, said at The AI Enterprise Conference 2026 in a panel on scaling AI agents past the pilot.

What should legal and risk leaders do now?

For general counsel and law firm technology leaders, the launch is a prompt to sharpen requirements rather than switch tools:

• Ask how ethical walls, matter permissions and client instructions are enforced, and whether they are auditable.
• Confirm data retention and human review terms in writing for each product tier.
• Build a small evaluation set from your own matters and compare tools before the API arrives.

The next milestone is the API release and its pricing, which will shape what vendors such as Harvey and Legora can offer. Until then, Astra for Law is a limited program, and its benchmark claims are OpenAI's own.

Frequently asked questions

What is OpenAI's Astra for Law?

A configuration of GPT-6 Astra introduced September 17, 2026, that adds a legal search index covering U.S. case law, statutes, regulations, court rules and administrative decisions, plus legal-analysis instructions. It is available first to selected law firms through OpenAI's Trusted Access program.

How accurate is Astra for Law?

On 200 questions from the private validation set of Vals AI's Legal Research Bench, OpenAI reported a 54.0% pass rate on its overall correctness check, versus 38.7% for GPT-6 Astra with web search. These are OpenAI's own results on a small sample.

Can legal software vendors build on Astra for Law?

OpenAI plans an API version called gpt-6-astra-law and named Harvey and Legora as customers, but has not announced a release date or pricing.

Filed under: AI Use Cases & Applications, AI Governance & Trust, Context, Semantics & RAG