OpenAI Launches Astra for Law but Keeps API Access Closed
A.I. / news
OpenAI Launches Astra for Law but Keeps API Access Closed
The legal-focused model searches more than 230 million U.S. legal sources, but the Sept. 17 rollout reaches only a small group of law firms so far.

OpenAI introduced Astra for Law on Sept. 17, a configuration of GPT-6 Astra tuned for legal research, but limited the release to a Trusted Access program covering a small number of Am Law 200 firms.
The product pairs GPT-6 Astra with a legal search index covering more than 230 million U.S. case law, statute, regulation, court rule and administrative-decision sources, according to reports by LawSites and Legal IT Insider, which both reviewed OpenAI's announcement.
What Astra for Law actually does
OpenAI built 26 partner plugins connecting the product to outside legal tools, including Thomson Reuters, Harvey, Legora, iManage, Intapp and DeepJudge, LawSites reported. OpenAI said the product answered 54 percent of 200 U.S. legal research questions correctly, up from 38.7 percent for GPT-6 Astra using web search alone, a comparison The Terminal could not independently verify. PYMNTS reported that same result as a 40 percent relative improvement, alongside a 24 percent increase in relevant reference cases found and a 54 percent increase in relevant court opinion passages retrieved.
| Partner | Category |
|---|---|
| Thomson Reuters | Legal research content |
| Harvey | Legal AI platform |
| iManage | Document management |
| DeepJudge | Legal search |
Who gets it first, and who doesn't yet
Access runs through Trusted Access, available in ChatGPT and Codex to the selected firms, with a zero-data-retention option for participants, PYMNTS reported. General API access is "coming soon," PYMNTS said, citing OpenAI, without a date, and PYMNTS reported that pricing for either tier has not been disclosed. Sullivan & Cromwell partner John Savva, Latham & Watkins AI strategy committee chair Michael Rubin and Cooley capital markets co-chair Dave Peinsipp were among the lawyers Legal IT Insider quoted reacting to early access.
The named hire running it
Jason Boehmig, head of OpenAI's legal vertical, joined the company in June after stepping down as chief executive of Ironclad, the contract-management startup he co-founded and built to a $3.2 billion valuation, according to Legal IT Insider. Sherwin Wu, a member of OpenAI's technical staff, led the product's engineering, LawSites reported.
What OpenAI did not benchmark
Legal IT Insider reported there is "no benchmark available comparing it with leading legal research providers," meaning OpenAI's published numbers measure Astra for Law only against its own base model, not against Westlaw, Lexis or the other incumbent tools its own partner firms already license. OpenAI has described the new product as complementing Thomson Reuters' licensed content rather than replacing it, the outlet said.
The launch follows a summer in which GPT-6 Astra completed 60 of 100 robot-arm commands a safety benchmark flagged as dangerous and a prior-art dispute over a rival decision model played out in public, both signs OpenAI is pushing GPT-6 Astra into narrower, higher-stakes product lines faster than outside evaluation has kept pace. OpenAI has not said when general API access will reach firms outside the Trusted Access program, or whether the legal index will ever expand beyond U.S. sources.
Sources
More in A.I.
- 01Meta Prices Muse Ember at $0.10 per Million Input Tokens, Says Nothing on WeightsMeta's smallest model scores 36 on Artificial Analysis, level with Claude Haiku 5.5, but it trailed Haiku on one of the two coding tests the firm ran.
- 02Anthropic Sets $70 to $78 Range for Nasdaq IPO, Crunchbase ReportsAn amended filing reportedly adds a price range and a $26 billion private placement, and shows a nine-month net loss of $104 billion on $33.2 billion in revenue.
- 03Google's Gemini Help Page Says Human Reviewers See Drive and Gmail ExcerptsA GitHub post from a user named thisisme6598 claims Gemini reads files unprompted. Google's own page confirms review, but not that part.
- 04Anthropic Ships Claude Fable 5.2 at $8 per Million Input TokensArtificial Analysis ranks the model first of 74 on its index, but its run needed 275 million output tokens and 44 minutes per task.