OpenAI Launches Astra For Law With GPT-6 logo

OpenAI Launches Astra For Law With GPT-6

OpenAI launched Astra for Law, a GPT-6-powered legal research tool searching across more than 230 million case-law and regulatory sources, available to select US law firms including Harvey and Legora as API customers.

By the Numbers

230M+ URLs
Sources indexed
54%
Benchmark pass rate
38.7%
vs. web search
26
Partner plugins
TC
By the AI Desk
Edited by Trace Cohen · Early-stage VC & angel · Founder, New York Venture Partners
2 min read
ShareXLinkedInEmail

THE RUNDOWN

1

Astra for Law's index spans more than 230 million URLs of US case law, statutes, regulations, court rules and administrative decisions, sourced substantially from the Free Law Project's CourtListener, which covers over 99.9% of published precedential case law -- a serious data-coverage claim, not a thin search wrapper.

2

On OpenAI's own benchmark, Astra for Law passed 54% of research questions versus 38.7% for GPT-6 Astra using plain web search -- a measurable, if self-reported, capability delta specific to legal research rather than a general model upgrade.

3

Existing legal-AI vendors Harvey and Legora are positioned as API customers building on top of Astra for Law, rather than being displaced outright -- OpenAI is competing at the model and data-infrastructure layer while leaving law firms' existing application relationships largely intact for now.

4

Access is limited to selected US firms through a "Trusted Access" program, with forward-deployed engineering work alongside firms including Sullivan & Cromwell, Ropes & Gray and Cooley -- a high-touch enterprise rollout rather than a self-serve product launch.

TC

The VC Read · Trace's Take

Trace Cohen

OpenAI selling the model layer to Harvey and Legora rather than launching a competing consumer product is the tell -- it's capturing the infrastructure economics while leaving the firm-relationship and workflow layer to vendors who already own it. The diligence item for anyone backing a legal-AI application startup: how much of your differentiation survives once the underlying model itself can pass 54% of the same benchmark you're selling against.

Analysis

OpenAI launched Astra for Law on September 17, a GPT-6-powered legal research product searching across more than 230 million case-law and regulatory sources, according to Artificial Lawyer and Legal IT Insider. The tool is available to selected US law firms through OpenAI's "Trusted Access" program.

What The Index Actually Covers

Astra for Law's search spans US case law, statutes, regulations, court rules and administrative decisions across more than 230 million URLs, with new sources added daily. Much of the case-law coverage comes from the Free Law Project's CourtListener, which the company says covers more than 99.9% of published precedential US case law -- a data-coverage claim that, if accurate, puts Astra for Law's underlying corpus close to parity with the specialized legal-research databases that have anchored the industry for decades.

A Measurable Benchmark Claim

OpenAI reported that Astra for Law passed 54% of research questions on its own internal benchmark, compared with 38.7% for the general-purpose GPT-6 Astra model using plain web search -- a specific, testable delta rather than a vague capability claim, though it's OpenAI's own benchmark and hasn't been independently replicated by a third party.

Competing With, Not Against, Legal-AI Incumbents

Notably, Harvey and Legora -- two of the best-funded legal-AI startups, both already embedded at large firms -- are positioned as API customers who will build on top of Astra for Law rather than as displaced competitors. That's a different posture than OpenAI's approach in some other verticals: rather than launching a consumer-facing legal product to compete head-on with Harvey and Legora's existing law-firm relationships, OpenAI is selling the underlying model and data layer to the same companies that already have those distribution relationships. The rollout includes 26 partner plugins and dedicated forward-deployed engineering work with major firms including Sullivan & Cromwell, Ropes & Gray and Cooley -- a high-touch enterprise sales motion closer to how OpenAI has approached other regulated-industry launches than a self-serve API release.

Why This Matters For The Broader AI Market

Legal research is one of the clearest enterprise categories where hallucination risk carries direct professional and malpractice consequences, making accuracy benchmarks like the 54%-versus-38.7% figure more consequential than a similar delta would be in a lower-stakes consumer use case. A frontier lab building vertical-specific retrieval and benchmarking infrastructure for law, rather than just prompting a general model with legal questions, signals OpenAI sees enough revenue potential in professional-services verticals to justify dedicated engineering investment beyond the core model.

The risk for law firms and legal-AI startups alike: if OpenAI's underlying model and data layer keeps improving faster than application-layer differentiation can be built on top of it, the value captured by Harvey, Legora and similar vendors could compress toward the interface and workflow layer, with OpenAI capturing more of the underlying economics. Whether that plays out depends on how defensible Harvey and Legora's own data, workflow integrations and firm relationships prove to be against a model provider that now offers comparable core research capability directly.

ShareXLinkedInEmail

More on

OpenAI

Key Sources

2 sources

THE WIRE in your inbox— Tech, startup & VC news with Trace's take. Free, no spam.