Law report No. GLW-1681 · filed November 19, 2026

Legal PracticeReported case

OpenAI Opens Astra for Law to All ChatGPT Enterprise Customers

OpenAI has opened Astra for Law, now built on GPT-6.2, to all ChatGPT Enterprise customers on request. It topped Harvey's LAB-AA benchmark at 20.8% and grew its legal index past 310 million URLs.

By Grace Kim4 min read800 words

Holding

  1. Astra for Law scored 20.8% on LAB-AA, versus 15.0% for base GPT-6.2 Astra and 13.1% for Muse Wonder Full.
  2. The legal search index grew 34% to more than 310 million URLs, covering 99.9% of published U.S. precedential case law.
  3. Astra for Law averaged 0.03 material hallucinations per assignment; Claude Opus 5.5 averaged 0.59.
  4. Access is now open to any ChatGPT Enterprise customer via request, plus API access.
  5. Vals AI Legal Research Bench score: 112 of 208 questions, or 53.8%, up from about 46% in September.
OpenAI Opens Astra for Law to All ChatGPT Enterprise Customers and Releases New Benchmark Results
PlateOpenAI Opens Astra for Law to All ChatGPT Enterprise Customers and Releases New Benchmark Results — Magic Fan / unsplash

OpenAI has opened Astra for Law to every ChatGPT Enterprise customer on a request basis, ending the Trusted Access-only rollout that in September limited the legal product to a handful of firms including Cooley, Ropes & Gray, and Sullivan & Cromwell. The updated release, announced this week, moves the product onto OpenAI's newest GPT-6.2 Astra base model and pairs it with an enlarged legal search index.

The company also claims benchmark leadership. OpenAI said Astra for Law achieved the top score on LAB-AA, the Artificial Analysis version of Harvey's Legal Agent Benchmark, and ranks as a top performer on the Vals AI Legal Research Bench.

What do the benchmark numbers show?

LAB-AA tests models on 120 assignments across 24 areas of law, from drafting deposition outlines to reviewing commercial leases. Scoring is all or nothing. An answer earns credit only if it meets every requirement without a material hallucination.

Astra for Law earned full credit on 20.8% of assignments. The base GPT-6.2 Astra model, not configured for law, scored 15.0%. Muse Wonder Full, the highest-scoring model from another developer, scored 13.1%.

The blog OrcaRouter noted the practical meaning of that gap: "It is the difference between finishing one in five legal assignments cleanly and one in seven." Cost also differs. Astra for Law used roughly 64,000 tokens per assignment, against about 139,000 for Muse Wonder Full — less than half the output for a higher score.

On hallucinations, Artificial Analysis counted an average of 0.03 material hallucinations per assignment for Astra for Law, tied for the lowest rate measured. Anthropic's Claude Opus 5.5, which OpenAI said had the best result among other developers, averaged 0.59.

On Harvey's own LAB, a separate benchmark, the share of assignments meeting every requirement rose from 4.6% with GPT-6 Astra to 14.6% with GPT-6.2 Astra, and to 16.2% with the legal configuration — meaning most of the gain came from the base model, with the legal setup adding 1.6 percentage points.

On the Vals AI Legal Research Bench, where practicing lawyers set questions on U.S. federal and state law, Astra for Law earned full credit on 112 of 208 questions, or 53.8%, up from about 46% in September. Notably, that 46% baseline reported this week differs from the 54% of 200 questions OpenAI reported at the September launch.

Julio Pereyra, head of legal research at Harvey, said in a statement published by OpenAI: "Astra for Law 6.2 is a material improvement over prior Astra models across our benchmarks." He added that pairwise judges preferred its briefs to those of GPT-6 Astra nearly 90% of the time, "without regression to prior models' accuracy."

How large is the legal search index?

At the September launch, the index covered more than 230 million URLs of U.S. case law, statutes, regulations, court rules and administrative decisions, drawing on the CourtListener database from the nonprofit Free Law Project. OpenAI has since grown it by 34%, adding more than 80 million URLs, including material from the U.K., Australia and Canada.

The index now spans more than 310 million URLs and covers more than 99.9% of published U.S. precedential case law, the company said. The release also adds instructions for researching a question, weighing authorities and writing an answer.

What does this change for practitioners?

Any ChatGPT Enterprise customer can now request access through the Trusted Access program in the admin console, and customers can request API access to build the model into their own tools. Firms weighing a move beyond pilot programs can now benchmark the product themselves rather than rely on vendor-reported scores, and the token-efficiency figures give procurement teams a concrete cost comparison.

Jason Boehmig, the former Ironclad CEO who leads OpenAI's legal products, has consistently stressed that Astra for Law is not a new model but a configuration of one, with legal-specific instructions, tooling and a search index layered on top.

Where does the frontier-model race stand?

Since May, every major frontier developer has moved into legal:

  • Anthropic launched Claude for Legal in May.
  • Google launched Gemini Enterprise for Legal in August.
  • Microsoft launched Copilot for Legal in October.
  • OpenAI launched Astra for Law in September and updated it this week.

This week's update focused almost entirely on model performance. It said nothing about new vendor and community plugins or custom skills, which were part of the September launch. It also made no mention of access to justice, which Boehmig called a "paramount concern" in September.

OpenAI reprised its launch theme in the announcement: "We're only at the beginning. We can't wait to see where you take the profession."

via lawnext.com (Source)

Filed under

  • legal-ai
  • astra-for-law
  • chatgpt-enterprise
  • legal-research
  • ai-benchmarks
Share this article:

More from Grace Kim

Grace Kim

Show full bio

Correspondent covering companies, deals and strategy in International law and legal practice at Global Law Wire.

336 articles

Also before the court

« Previous articleNext article »