OpenAI Launches Astra for Law with a 230-Million-URL U.S. Legal Index

What was announced — and what still needs checking

From a general model to legal infrastructure

The launch is more than a model preset. OpenAI combines GPT-6 Astra with a legal search index, analysis and drafting instructions, and specialist tools so firms and legal-technology vendors can build workflows around their own methods and review standards.

The examples span authority research, multi-document contract analysis, diligence, and drafting advice that can be traced back to source material. The intended division of labour is clear: AI handles retrieval and synthesis; qualified professionals remain responsible for judgment.

What the 230-million-URL index means

OpenAI says the index covers more than 230 million U.S. legal URLs across case law, statutes, regulations, court rules, and administrative decisions, with sources added daily. Its Free Law Project integration includes a collection covering more than 99.9% of published U.S. precedential case law.

Coverage does not eliminate legal-research risk. Users still have to check whether authority remains good law, whether it is binding in the relevant jurisdiction, and whether the facts are sufficiently analogous—especially outside U.S. law.

A stronger benchmark result, with narrow conditions

At the highest reasoning effort, OpenAI reports that Astra for Law passed the overall-correctness check on 54.0% of 200 questions from a private Vals AI Legal Research Bench validation set, versus 38.7% for GPT-6 Astra with web search. The stated 40% improvement is relative, not a 40-point increase.

OpenAI also reports 24% more reference cases on case-law questions and up to 54% more relevant passages from the correct opinions. These figures come from the developer and do not establish comparable performance for Thai law, Thai-language records, or an organization's own matters.

Access, data controls, and unresolved terms

GPT-6 Astra Law will appear in the model picker for selected firms accepted into Trusted Access. The API model is named gpt-6-astra-law, but OpenAI has not provided a general launch date or pricing.

For eligible firms, OpenAI describes Zero Data Retention on the API and says ChatGPT Enterprise use is excluded from human review by default. Those are distinct product surfaces and should not be generalized to every account, plugin, or partner service without checking the applicable contract.

An ecosystem that also expands the risk surface

The launch includes 26 partner-built plugins plus nine community plugins and 47 adaptable custom skills. Examples cover matter-file storage, precedent comparisons, time-entry review, and importing specialist context into ChatGPT.

Each connection adds a data path. Security and legal teams should verify the publisher, OAuth scopes, token handling, logging, data location, revocation, and matter-level segregation before confidential client information is introduced.

What Thai organizations can do now

The immediate lesson for Thailand is architectural: trustworthy retrieval, permissions, and review workflows matter as much as the base model. The announcement does not confirm a Royal Gazette corpus, Thai judgments, or a Thai-language legal evaluation.

A sensible pilot would start with internal documents that have reference answers—such as extracting clauses or comparing contract versions—while prohibiting final legal conclusions and requiring professional review of jurisdiction, authority, citations, and facts.

Announcement and source status

OpenAI announced Astra for Law on 17 September 2026, alongside Trusted Access and the general availability of ChatGPT for Word. The announcement page does not show a separate modification date. OpenAI's supporting Enterprise privacy page states that it was updated on 8 January 2026.

Frequently asked questions