OpenAI Astra for Law launches with GPT-6 Astra, a legal search index spanning more than 230 million URLs and specialized instructions for legal analysis and writing. The new product targets law firms and legal-technology companies rather than consumers seeking automated legal advice.

 

OpenAI will initially provide the system to selected firms through Trusted Access in ChatGPT and Codex, with an API version planned. The launch pushes the company into direct competition for professional legal workflows while preserving lawyers’ responsibility to verify authorities, protect client confidences and exercise independent judgment.

 

OpenAI’s September 17 release has four defining elements:

  • A U.S. legal index covering more than 230 million URLs
  • A 54% correctness result on a private legal-research test set
  • Twenty-six plugins connecting established legal platforms
  • Confidentiality controls for eligible law firms

 

OpenAI Astra for Law Searches 230 Million URLs

OpenAI says Astra for Law combines its most capable model with a dedicated index covering U.S. case law, statutes, regulations, court rules and administrative decisions. Sources are added daily, giving the system a more structured legal corpus than general web search alone.

 

The index includes material from Free Law Project’s CourtListener collection, which covers more than 99.9% of published U.S. precedential case law. OpenAI positions that public-law layer as a complement to licensed databases and specialist products, including services from Thomson Reuters.

 

Custom instructions guide the model to distinguish binding holdings from other judicial observations, identify contrary authorities and explain uncertainty. The same configuration can apply research to a client’s facts, proposed deal terms or contractual risk, but its output remains a starting point for professional review.

 

The 54% Legal Research Score Still Requires Lawyer Review

OpenAI tested the complete Astra for Law configuration on 200 U.S. questions from a private validation set of Vals AI’s Legal Research Bench. At the highest reasoning setting, it passed the benchmark’s overall correctness check on 54% of questions.

 

GPT-6 Astra using ordinary web search passed 38.7% under the same conditions, producing the reported 40% relative improvement. On case-law questions, Astra for Law found 24% more reference cases and retrieved up to 54% more relevant passages from the correct opinions.

 

The benchmark supports OpenAI’s claim that legal indexing improves its base model, but it does not establish error-free research. A 54% strict correctness rate means a substantial share of answers still missed at least one required criterion. OpenAI also did not publish a direct benchmark against major licensed legal-research platforms.

 

That limitation is especially important in law, where an invented citation, outdated authority or missed jurisdictional distinction can affect a filing or client decision. The product’s value will depend on whether lawyers can inspect every cited passage and trace conclusions back to authoritative sources.

 

Related Coverage

 

Trusted Access Adds Confidentiality and Governance Controls

Astra for Law will appear in ChatGPT and Codex as “GPT-6 Astra Law” for eligible firms. OpenAI says the coming API identifier will be gpt-6-astra-law, allowing legal software companies such as Harvey and Legora to build the configuration into their own products.

 

The Trusted Access program includes zero data retention for eligible API use. ChatGPT Enterprise activity under the offering is excluded from human review by default. OpenAI is also working with Latham & Watkins on permissions, ethical walls, client instructions and firm-level oversight.

 

Those controls address a central barrier to legal AI adoption: confidential matter data cannot be treated like ordinary consumer prompts. Firms will still need to test how access rules, audit trails and client restrictions behave across every connected product and internal knowledge source.

 

OpenAI identified Sullivan & Cromwell, Ropes & Gray, Cooley, Latham & Watkins and Wachtell Lipton among firms involved in testing or developing applications. Reuters independently confirmed the initial limited-access strategy and the product’s legal-market positioning.

 

Twenty-Six Plugins Turn the Model Into a Legal Platform

OpenAI is launching 26 ecosystem plugins linking ChatGPT with legal services such as Clio, DeepJudge, iManage, Intapp, Relativity and Thomson Reuters. The integrations are designed to keep documents, matter context and firm knowledge inside existing permission systems while making them available during AI-assisted work.

 

Early firm projects show how the platform can be customized. Sullivan & Cromwell built an agreement analyzer using negotiating playbooks and precedents. Ropes & Gray developed a deal-diligence workflow, while Cooley’s GO Public system supports IPO preparation and risk review.

 

The strategy is broader than selling a specialized chatbot. OpenAI wants its model and interface to become a foundation on which firms, research providers and legal-software vendors build proprietary workflows. That ecosystem approach lets the company enter professional markets without replacing every established database or matter-management system.

 

Competition is already intense. Google has expanded Gemini Enterprise tools for legal users, Anthropic offers Claude capabilities for lawyers, and Thomson Reuters has introduced its own professional model backed by proprietary content. Astra for Law’s adoption will be judged on citation reliability, integration quality and governance—not model fluency alone.

 

For law firms, the immediate opportunity is faster research, diligence and drafting with traceable sources. The unresolved question is whether the system can maintain accuracy across real matters that are more ambiguous than benchmark questions. OpenAI’s limited rollout gives firms a controlled setting to answer that before wider API availability.