OpenAI has introduced Astra for Law, a new foundation that combines its latest GPT-6 Astra model with a dedicated legal search index and custom instructions for legal analysis and writing. The offering targets law firms and legal technology companies, with API access rolling out to partners including Harvey and Legora. The model will appear as "GPT-6 Astra Law" in the ChatGPT model picker and as gpt-6-astra-law in the API.
The legal search index covers U.S. case law, statutes, regulations, court rules, and administrative decisions across more than 230 million URLs, with sources added daily. OpenAI says the index incorporates CourtListener's case-law collection from the Free Law Project, covering more than 99.9% of published U.S. precedential case law. On Vals AI's Legal Research Bench validation set, Astra for Law passed the overall correctness check on 54.0% of questions at the highest reasoning effort, compared with 38.7% for GPT-6 Astra using web search alone — a 40% relative improvement. On case-law-focused questions, it found 24% more reference cases and retrieved up to 54% more relevant passages from correct court opinions.
OpenAI is also expanding privacy and governance controls for confidential client work. Eligible firms get Zero Data Retention on the API, and ChatGPT Enterprise usage is excluded from human review by default. The company is working with Latham & Watkins to design information permissions, ethical walls, client instructions, and firm oversight. ChatGPT for Word is now generally available, letting lawyers proofread, get suggested edits, and flag formatting issues directly in Word.
Confirmed
- Astra for Law combines GPT-6 Astra with a legal search index and custom instructions for legal analysis and writing.
- The legal search index searches over 230 million URLs covering U.S. case law, statutes, regulations, court rules, and administrative decisions.
- CourtListener data from the Free Law Project covers more than 99.9% of published U.S. precedential case law.
- On Vals AI Legal Research Bench: 54.0% overall correctness at highest reasoning effort vs. 38.7% for GPT-6 Astra with web search (40% relative improvement).
- 24% more reference cases found on case-law questions; up to 54% more relevant passages retrieved from correct opinions.
- Initial access via Trusted Access Program in ChatGPT and Codex; API access coming soon as
gpt-6-astra-law. - Zero Data Retention (ZDR) on API for eligible firms; ChatGPT Enterprise usage excluded from human review by default.
- 26 partner plugins launched covering practice and business of law (iManage, Intapp, DeepJudge, Thomson HighQ/CoCounsel preview, plus 9 community plugins with 47 custom skills).
- ChatGPT for Word now generally available.
- Named firm collaborations: Sullivan & Cromwell (agreement analyzer), Ropes & Gray (deal diligence system), Cooley (GO Public for IPO prep), Latham & Watkins (governance design), Wachtell Lipton Rosen & Katz (litigation and corporate expertise).
Unknown
- Pricing for API access (
gpt-6-astra-law) and Trusted Access Program tiers. - Exact rollout timeline for general API availability beyond "coming soon."
- Eligibility criteria and application process for the Trusted Access Program.
- Independent replication of Vals AI benchmark results by third parties.
- Whether the legal search index covers non-U.S. jurisdictions or will expand beyond U.S. law.
- Specific SLA, uptime, or data residency guarantees for law firm deployments.
- Details on how "ethical walls" and information permissions designed with Latham & Watkins will be technically enforced.
Our take
OpenAI is selling infrastructure, not a finished legal app — firms get the model, search index, and plugin ecosystem to build their own workflows. The 40% relative benchmark gain is notable but on a private validation set; independent replication will drive adoption. The composable approach lets firms mix OpenAI's frontier model with specialist tools and their own precedents without vendor lock-in. The Trusted Access Program with ZDR addresses the confidentiality blocker that has kept many large firms from putting client data into frontier models.
