OpenAI Astra for Law Launches with 230-Million-URL Index, Targeting the Legal Research Stack
OpenAI launched OpenAI Astra for Law on September 17, 2026, pairing its most powerful model, GPT-6 Astra, with a 230-million-URL index of US legal material and a governance package designed for law firms. The product debuts inside ChatGPT and Codex through a Trusted Access programme, with the API following later as gpt-6-astra-law.
The real story is the corpus. OpenAI’s legal search index covers case law, statutes, regulations, court rules and administrative decisions across the United States, with case law sourced from the Free Law Project’s CourtListener nonprofit repository. According to OpenAI, that collection reaches 99.9 percent of published US precedential case law. The index is the foundation that lets Astra for Law answer narrow jurisdiction-specific questions rather than rely on the open web, and it determines which firms can trust its citations on day one. ChatGPT for Word also became generally available on September 17, giving lawyers a surface where the index and the model meet the document they are drafting.
Performance numbers came from the Vals AI Legal Research Bench private validation set, a benchmark of 200 US legal research questions. At its highest reasoning effort, Astra for Law passed an overall correctness check on 54.0 percent of the questions. The same model, using web search alone and no legal index, passed 38.7 percent. OpenAI characterised that 15.3-percentage-point gap as a 40 percent relative improvement. On case-law questions specifically, the company said Astra for Law surfaced 24 percent more reference cases and retrieved up to 54 percent more relevant passages from correct opinions. OpenAI published a side-by-side comparison against Anthropic’s Claude Fable 5.1 as part of the same release.
OpenAI Astra for Law Arrives With 26 Partner Plugins and Governance Stack
The launch is built around 26 partner-built plugins that pipe firm systems into the model. Relativity, Clio, iManage, Intapp and DeepJudge are among the named integrations, with Thomson Reuters contributing HighQ matter context and a CoCounsel Legal connector. Nine additional community plugins come from LegalQuants, LECG and Skills.law, shipping with 47 custom skills for document review, citation checking and transactional workflows. The pattern matters: Astra for Law is positioned less as a chatbot and more as a governance surface that can read, write and audit inside the software lawyers already run.
Two of the legal AI sector’s most valuable independents are building on top of OpenAI infrastructure. Harvey, valued at $15.6 billion in a financing reported earlier this month, is constructing products through the Astra API, while Legora, which is reported to be pursuing a $10 billion valuation, is doing the same. Their bet is that the underlying model and index can be a shared commodity while their own interfaces, matter context and pricing carry the differentiation. That puts pressure on firms that built earlier stacks on different base models.
OpenAI Astra for Law Zero Data Retention and Per-Firm Controls
Governance features were foregrounded in the announcement. Trusted Access firms get Zero Data Retention on the API by default, and ChatGPT Enterprise usage is excluded from human review by default, with controls that legal operations leaders can override per matter. OpenAI said it is working with Latham & Watkins on information permissions, ethical walls, client instructions and firm oversight, the kind of granular controls large law firms demand before allowing client material to leave their perimeter. Six vendors issued coordinated releases between 16:00 and 16:15 New York time on launch day: iManage, Relativity, Consilio, Litera, Laurel and Everlaw, each announcing compatibility or connector updates timed to the model release.
Several named firms have already built on the platform. Sullivan and Cromwell constructed an agreement analyser, Ropes and Gray built a diligence system and Cooley produced a tool called GO Public for IPO preparation. Wachtell Lipton Rosen Katz is using the system for litigation and corporate matters. OpenAI framed these as proof points that the model can be deployed against high-stakes transactional and contentious work without rebuilding firm-specific pipelines from scratch. The OpenAI Astra for Law launch marks a structural shift in how frontier AI providers package vertical-specific offerings for regulated industries, and OpenAI Astra for Law is positioned as the template other industries will receive.
The launch continues a vertical pattern at OpenAI. A week earlier, on September 10, 2026, the company released ChatGPT for Financial Services, connecting GPT-6 Astra to data from S&P, LSEG, Moody’s and PitchBook. The legal product mirrors that architecture: a domain-tuned model, licensed third-party data, enterprise governance defaults and an API layer for partners. It also lands in a market Google entered in August with Gemini Enterprise for Legal, a product built with Freshfields, Cleary Gottlieb and Weil, Gotshal & Manges. Two frontier model providers are now selling curated legal stacks with named-firm reference customers, and the question for general counsel is which index, which governance defaults and which partner ecosystem they want inside their matter workflow.
Source: TheNextWeb