OpenAI launches Astra for Law, and its own legal research index

“As AI becomes more open and interoperable, the value is not in connectivity alone,” said Joel Hron, chief technology officer of Thomson Reuters. “Legal professionals need more than access to information.”

He said it in a statement supporting OpenAI, on the day OpenAI started building its own legal research index.

OpenAI introduced Astra for Law on 17 September. It is GPT-6 Astra, the company’s most powerful model, configured with legal search, instructions for legal analysis and writing, and controls aimed at law firms. Selected firms get it first through a Trusted Access programme in ChatGPT and Codex. The API follows later, as gpt-6-astra-law.

The index is the news

The model configuration matters less than what sits underneath it. OpenAI has built a legal search index covering United States case law, statutes, regulations, court rules and administrative decisions. It spans more than 230 million URLs, with sources added daily.

Much of the case law comes from the Free Law Project, the nonprofit behind CourtListener. OpenAI says that collection covers more than 99.9% of published US precedential case law.

The company frames this as complementary. The index “complements the licensed content and specialist products firms rely on from providers such as Thomson Reuters”, its announcement says. Thomson Reuters owns Westlaw, sells CoCounsel, and has spent this year building its own legal models. Its CTO supplied a supportive quote for the launch.

What the benchmark shows, and what it does not

OpenAI tested the configuration on 200 US legal research questions from the private validation set of Vals AI’s Legal Research Bench. At the highest reasoning effort, Astra for Law passed the overall correctness check on 54.0% of questions. GPT-6 Astra with web search alone passed 38.7%.

OpenAI presents that as a 40% relative improvement, which it is. It also means the configured system fails the correctness check on nearly half the questions. On case-law questions it found 24% more reference cases, and retrieved up to 54% more relevant passages from the correct opinions.

The testing is OpenAI’s own. So is the side-by-side comparison it published against Claude Fable 5.1, in which OpenAI says the rival model cited a holding an appeal court had overturned, and in a second example reported finding no relevant case at all. No independent party has audited either result.

Twenty-six plugins, and six press releases

Alongside the model, OpenAI launched 26 partner-built plugins connecting ChatGPT to tools firms already use, including Relativity, Clio, iManage, Intapp and DeepJudge. Thomson Reuters is bringing HighQ matter context into ChatGPT and previewing a CoCounsel Legal connector. Nine community plugins, from LegalQuants, LECG and Skills.law, arrive with 47 custom skills. ChatGPT for Word became generally available the same day.

The vendors announced themselves, in formation. Six of them issued releases between 16:00 and 16:15 New York time: iManage, Relativity, Consilio, Litera, Laurel and Everlaw. Everlaw went furthest, letting lawyers query the substance of case evidence rather than metadata about it. Five carried the same quote from Jason Boehmig, OpenAI’s general manager for the legal industry, word for word.

Read together, those releases draw one line repeatedly. Time entry continues in Laurel. Deeper workflows continue in Litera. Relativity keeps its substantive analysis inside its own platform. Consilio passes only material responsive to the question asked.

Each vendor is describing governance. Each is also describing where the subscription stays, now that the question gets asked somewhere else.

Harvey and Legora are building on it too

Harvey, valued at $15.6bn this month, will build on Astra for Law through the API. So will Legora, which has been chasing a $10bn valuation. Harvey’s head of applied research, Niko Grupen, praised its citation precision in a statement OpenAI published.

Those two are the best-funded companies selling legal AI as a destination. They will now build on infrastructure sold by a company that also ships legal research directly to law firms.

OpenAI says that is the point. “Building on OpenAI should mean getting more from the ecosystem, not replacing it,” the announcement reads.

The governance package

Law firms have been slow for a reason, and OpenAI addressed it directly. Eligible firms in the Trusted Access programme get Zero Data Retention on the API. OpenAI excludes ChatGPT Enterprise usage from human review by default.

OpenAI is also working with Latham & Watkins on information permissions, ethical walls, client instructions and firm oversight. “As AI becomes more capable, so too does the ability to deploy it in environments that demand rigorous governance, oversight, and accountability,” said Michael Rubin, chair of Latham’s AI strategy committee.

OpenAI has put its forward-deployed engineers inside firms too. Sullivan and Cromwell built an agreement analyser that brings its negotiating playbooks into deal review. Ropes and Gray built a diligence system for working through a data room. Cooley built GO Public for IPO preparation. Wachtell, Lipton, Rosen and Katz is working with OpenAI on litigation and corporate matters.

The same move, one vertical over

OpenAI did not improvise any of this. A week earlier, OpenAI launched ChatGPT for Financial Services with data from S&P, LSEG, Moody’s and PitchBook. Frontier model, licensed or indexed data, a governance wrapper, partner connectors. Law is the same template with different filings.

Google is running its own version. It launched Gemini Enterprise for Legal in August with Freshfields, Cleary and Weil.

What OpenAI has not said is the price, which firms are inside Trusted Access, or when any of this reaches Europe. The search index covers United States law only. For European firms the governance questions are sharper, and the announcement does not address them.

The composability line is doing a lot of work here. On one reading, OpenAI is handing the legal industry better tools and letting specialists keep their franchises. On another, it now owns the model, the research index, the distribution surface and the plugin directory, and the specialists own the filing cabinets. Both can be true for a while. Only one of them is stable.