OpenAI launches Astra for Law: GPT-6 Astra with search index of American legal materials

OpenAI launches Astra for Law: GPT-6 Astra with search index of American legal materials

OpenAI on Thursday launched Astra for Law, a legal configuration of GPT-6 Astra with its own search index over American legal material, access controls, and a Trusted Access program for confidential client work.

OpenAI on Thursday launched Astra for Law, a legal configuration of GPT-6 Astra with its own search index over American legal material, access controls, and a Trusted Access program for confidential client work. The launch comes just weeks after both Google and Thomson Reuters directed their own products at the same customer group — the legal profession has become an active competitive arena for the frontier labs.

What Was Launched

Astra for Law was introduced on September 17, 2026 and, according to OpenAI, is designed to help law firms and legal software vendors with research, draft advisory work, and the development of custom applications for lawyers. The product combines the GPT-6 Astra model with a search index over American case law, statutes, regulations, and other legal material, along with specialized instructions for legal analysis and writing. The index spans more than 230 million URLs, according to the company, and the package includes access controls and workflow tools built for law firms and legal technology companies.

How the Product Works

The core is the connection between the model and the index. OpenAI says that a collaboration with the Free Law Project, the nonprofit behind CourtListener, brings a case law collection covering more than 99.9 percent of published American precedent into the research experience. The company presents the index as a supplement to licensed content and specialist products firms already use from vendors such as Thomson Reuters — not as a replacement.

In addition to the model and index, the product consists of specialized instructions for legal analysis and writing, and access-management tools, so that firms can control who gets access to what.

Rollout: Trusted Access First, API Later

Astra for Law is initially offered to selected law firms through Trusted Access in ChatGPT and Codex, a program that, according to Reuters, includes protections for confidential client work. API access follows later: the product will appear in the model picker as GPT-6 Astra Law and in the API as gpt-6-astra-law. Neither pricing nor a date for the API access has been given.

OpenAI states that it has tested the product together with Sullivan & Cromwell, Ropes & Gray, Cooley, Latham & Watkins, and Wachtell Lipton. According to OpenAI, the company's forward-deployed engineers have customized ChatGPT Enterprise with bespoke interfaces and integrations against the firms' proprietary data. Among the examples: Sullivan & Cromwell has built a contract analyzer that pulls the firm's negotiation playbooks and selected precedent into its review of new agreements, to surface risks that arise when provisions are read together — and translate the findings into proposed redlines and drafts of client advice. Ropes & Gray has, according to the announcement, built a due diligence system for transactions, and Cooley has developed the IPO tool GO Public.

The Numbers — With Major Caveats

OpenAI reports that Astra for Law, at the highest reasoning effort for both systems, achieved 54.0 percent overall accuracy on Vals AI's Legal Research Bench, versus 38.7 percent for GPT-6 Astra with web search only — a figure the company describes as a 40 percent relative improvement.

It is important to emphasize what this means: the benchmark was run on a private validation set of 200 questions, the results are OpenAI self-assessments that have not been independently verified, and the comparison configuration — GPT-6 Astra with web search — was chosen by OpenAI itself. The number says something about what the company itself claims, not about an independently measured competitive advantage.

The Competitive Landscape

The launch marks, according to Reuters, the newest push by the frontier labs into the legal market. Google expanded its Gemini Enterprise offerings for legal professionals last month, and Anthropic has since January launched tools for lawyers built on the Claude assistant. In the same month as Google, Thomson Reuters launched Thomson 1.0 — a model from Thomson Reuters, the company behind Westlaw, trained on its own legal research content and built on open-source technology.

At the same time, Thomson Reuters is also an OpenAI partner: OpenAI says vendors such as Harvey and Legora can build products on Astra for Law, and that the platform will include integrations with, among others, Relativity, Clio, Intapp, and Thomson Reuters. The picture is thus double-edged: Thomson Reuters is both supplying its own model competing with OpenAI's offering and simultaneously a planned integration partner.

Open Questions

Several essential details are missing from the sources. The terms of the confidentiality protections in the Trusted Access program are not specified — for law firms, which carry liability exposure and professional duties of confidentiality, this is decisive. The same applies to pricing, the timing of API access, and how far the planned integrations with Relativity, Clio, Intapp, and Thomson Reuters have actually progressed. The partners' own statements about the product also cannot be fully verified from the available material.

What does come through is the shape of the move: OpenAI is offering not just a model but a whole package — model, case law index, instructions, access control, and distribution channel — in a market where Google, Anthropic, and Thomson Reuters have already positioned themselves. Who wins the legal profession will likely depend less on benchmark numbers like 54.0 versus 38.7 percent, and more on who can demonstrate confidentiality, source quality, and integration with the systems firms already use.