OpenAI Unveils Astra for Law with 230M URLs and 54% Accuracy

The GPT-6 Astra, configured for lawyers, correctly answered 54% of 200 questions from the Legal Research Bench developed by Vals AI, compared to 38.7% from the same model operating with conventional internet search. With these numbers, OpenAI has launched Astra for Law: a configuration of GPT-6 Astra embedded in a legal research index with over 230 million URLs of U.S. case law, statutes, regulations, and administrative decisions, updated daily. Initial access is restricted to the Trusted Access program within ChatGPT and Codex, with priority given to firms listed in the Am Law 200, ranking the largest U.S. law firms by revenue.

The Index and Its Contents

The document base has been built in partnership with the Free Law Project and covers over 99.9% of all published judicial precedents in the United States. Categories include federal and state case law, legislation, agency regulations, procedural rules, and administrative decisions. Latham & Watkins, Ropes & Gray, Cooley, Sullivan & Cromwell, and Wachtell Lipton participated in pre-launch testing. The product incorporates confidentiality controls and legal writing instructions that replace the generic configuration of the standard GPT-6 Astra, and comes with 26 plugins for tools like Relativity, Clio, Thomson Reuters, iManage, and DeepJudge.

Harvey Becomes a Partner, Not a Rival

The main market reading is the repositioning of Harvey. The legal AI startup, which raised $100 million in 2024 with participation from OpenAI itself, has built its products on OpenAI models and secured contracts with major U.S. and U.K. firms. With the launch of Astra for Law, the relationship shifts: Harvey and Legora appear among the 26 available plugins, repositioned as integration partners. OpenAI enters the legal market as a platform, not a final product, replicating the logic Salesforce used in CRM: capturing infrastructure revenue without displacing existing vertical partners who have already secured clients.

For firms already running Harvey on OpenAI APIs, Astra for Law adds a layer of document research without requiring immediate migration. Cost duplication is expected to be addressed in the next 12 to 18 months as contracts with vertical partners expire and firms assess whether the 40% accuracy gain in benchmarks justifies direct engagement with OpenAI.

The Perspective Beyond the United States

In the United Kingdom, Harvey and Legora already operate in large law firms. Their presence in the Astra for Law launch as integration partners, rather than replaced competitors, sends a clear signal about the distribution strategy: OpenAI prefers to grow through the network of already established partners in mature markets rather than engage in channel wars. For large U.K. firms, the central economic question becomes whether the markup charged by Harvey on OpenAI's infrastructure remains justified when the base product delivers 40% more accuracy in independent benchmarks than the previous alternative.

For firms in continental Europe and Brazil, Astra for Law arrives with limited utility for now. The index exclusively covers U.S. law. Firms practicing international law, bilateral arbitration with American counterparts, or regulatory monitoring of the FDA, SEC, and CFPB find immediate value in the product. For those operating under civil law jurisdictions without direct exposure to U.S. jurisdiction, the tool still does not meet local legal research needs.

Astra for Law is the first explicitly labeled vertical edition of GPT-6. Health, finance, and auditing have analogous datasets: vast volumes of regulated documents, public benchmarks with measurable results, and clients willing to pay for the automation of specialized knowledge work. The question that the market will raise in the coming months is whether the key partner in these verticals will continue to be the startup that built the layer of specialization, or if OpenAI will follow the same path it took in law: publishing niche infrastructure and transforming partners into distributors of its platform's product.