OpenAI released an updated Astra for Law on 17 November 2026 and opened it to every ChatGPT Enterprise customer, who can now request access through OpenAI’s Trusted Access Program, including via the API. The legal configuration of GPT-6.2 Astra ranked first on Artificial Analysis’ LAB-AA legal benchmark, according to OpenAI.
Quick Take
- What: a model configuration that pairs GPT-6.2 Astra with a legal search index and instructions for legal research and writing.
- Access: all ChatGPT Enterprise customers can request it via the admin console; API access by request form.
- LAB-AA: full credit on 20.8% of 120 legal assignments, ahead of GPT-6.2 Astra (15.0%) and Muse Wonder Full (13.1%).
- Hallucinations: 0.03 material hallucinations per assignment, versus 0.59 for Claude Opus 5.5, per OpenAI.
- Search index: grown 34% since September to more than 310 million URLs, now including the UK, Australia and Canada.
In plain English
Astra for Law is ChatGPT tuned for lawyers. It uses OpenAI’s newest model, adds a large database of court decisions, statutes and rules, and follows a lawyer-like method: research, weigh the authorities, then write. Until now only a few firms could use it. Now any company on ChatGPT Enterprise in eligible countries can ask for access, and only lawyers and people working under their supervision may use it.
What happened
OpenAI introduced Astra for Law in September to a small group of firms, including Cooley, Ropes & Gray and Sullivan & Cromwell. The new version runs on GPT-6.2 Astra, which launched the same day. “Today, we’re releasing an updated Astra for Law that is state-of-the-art on Harvey LAB-AA and a top performer on Vals AI’s Legal Research Bench,” OpenAI wrote. Harvey’s head of legal research, Julio Pereyra, is quoted saying the model “writes more comprehensive and better reasoned briefs that are preferred to those of GPT-6 Astra nearly 90% of the time by our pairwise judges.”
Background
Legal work is one of the most competitive enterprise AI markets, with specialists such as Harvey and model makers all chasing law firms. Accuracy is the main concern: a single invented citation can damage a lawyer’s credibility in court. That is why OpenAI leads with hallucination numbers alongside benchmark scores.
What’s in it
- LAB-AA: Artificial Analysis’ version of Harvey’s Legal Agent Benchmark, 120 assignments across 24 areas of law. Full credit requires meeting every requirement with no material hallucination.
- Efficiency: about 64,000 output tokens per assignment versus 139,000 for Muse Wonder Full, which OpenAI says matters for teams paying by usage.
- Vals AI Legal Research Bench: full credit on 112 of 208 questions (53.8%) in the 16 November evaluation, up from about 46% in September; 90.9% of available points including partial credit.
- Coverage: OpenAI says the index covers more than 99.9% of published US precedential case law.
- How to get it: a workspace admin verifies the business in the Trusted Access section of the admin console and signs a Legal Technology and Trust Addendum; approved teams then select gpt-6.2-astra-law in ChatGPT.
Why it matters
For law firms and in-house teams, broad access removes the waitlist, and lower token use can make long research tasks cheaper.
For legal AI startups, OpenAI is now a direct competitor on the research layer, not only a model supplier.
For buyers, the numbers are useful, but the best result is still full credit on about one in five complex assignments, so human review remains essential.
What’s still open
- Independent checks of the hallucination comparison, which comes from OpenAI’s reading of Artificial Analysis results.
- Pricing for API access and heavy enterprise use.
- How well the newer UK, Australian and Canadian coverage performs compared with US law.
Sources
- OpenAI, “Astra for Law leads in legal intelligence,” 17 November 2026
- OpenAI Help Center, “Astra for Law”
i10x Perspective
Model configurations like this are OpenAI’s way into regulated professions: same frontier model, plus a curated index and a method. Law is the test case; finance and medicine are the obvious next candidates. More on the OpenAI company page.






