OpenAI Astra for Law Launches: 230M Cases, 54% Research Accuracy (2026)
Astra for Law is OpenAI's legal-industry edition of GPT-6 Astra, pairing the model with a dedicated US legal search index. It shipped on September 17, 2026. Anyone who has hired a lawyer knows document review runs for days and bills by the hour. OpenAI's legal AI aims straight at that clock.
Why OpenAI's Legal AI Arrived Now
Law sat out the first AI wave for one reason: hallucination (when a model invents information that sounds plausible but does not exist). Cite a fake case and a filing collapses. Yet legal is also the market AI vendors want most, because at large firms research hours convert directly into billable revenue.
Astra for Law Benchmarks, By the Numbers
OpenAI published three figures. The Legal Search Index covers more than 230 million URLs of US case law, statutes, regulations and administrative decisions. On 200 US legal research questions, overall correctness reached 54.0% against 38.7% for GPT-6 Astra using web search alone, a 15.3 point jump and a 40% relative improvement. It also retrieved 24% more case references. Still, 54% is barely past half.
Why Six Elite US Firms Helped Build It
The design partner list reads like an Am Law roster: Sullivan & Cromwell, Ropes & Gray, Cooley, Skadden, Wachtell Lipton and Latham & Watkins. Access opens through Trusted Access in ChatGPT and Codex, with API availability coming as gpt-6-astra-law. Eligible firms get Zero Data Retention.
What This Means for the Legal AI Market
The competitive axis just moved. OpenAI shipped ChatGPT for Financial Services on September 10 and followed with law one week later, which makes this a deliberate vertical land grab. Thomson Reuters and LexisNexis built their moats on proprietary legal data and now face a model company on their own turf. Harvey and Legora, the best-funded legal AI startups, are building on top of Astra for Law rather than against it. Analysts expect model quality layered over data, not raw data volume, to settle this one.
Key Takeaways
① 230 million pages indexed - OpenAI wired GPT-6 Astra to a dedicated US case law and statute index.
② 38.7% to 54.0% accuracy - a 40% relative gain on a 200-question legal research benchmark.
③ Six elite firms on board - the AI race is shifting from general models to profession-specific ones.
Which profession a technology reaches first is not decided by capability alone. It follows the money on the table and how cleanly the data is already organized. After finance and law, the next queue is worth watching.
👉 GPT-6 Astra Launch: 'Critical' AI Declares AGI Era, Split Scorecard vs Claude - also worth a read.
📌 Sources: Reuters, OpenAI, CNBC (2026)
댓글
댓글 쓰기