OpenAI launches Astra for Law, with its own case-law index
But the index, not the model, is doing the work.
The facts
- Astra for Law pairs GPT-6 Astra with a legal search index and instructions for legal analysis and writing.
- The index covers American case law, statutes, regulations, court rules and administrative decisions across more than 230m URLs.
- OpenAI reports 54.0% overall correctness on 200 questions from Vals AI's Legal Research Bench, against 38.7% for GPT-6 Astra with web search alone.
- Harvey and Legora are named as API customers. Both sell legal AI products of their own.
OpenAI has launched Astra for Law, a version of GPT-6 Astra aimed at law firms. It is not a new model. The company has bundled its frontier model with a legal search index, instructions for legal analysis, and controls for confidential client work.
The index is the substantive part. OpenAI says it covers American case law, statutes, regulations, court rules and administrative decisions across more than 230m URLs, with sources added daily. Its case-law coverage arrives through the Free Law Project, the nonprofit behind CourtListener, which OpenAI says spans more than 99.9% of published American precedential case law. Thomson Reuters and other licensed providers are framed as complements, not casualties.
The company reports testing on 200 questions from the private validation set of Vals AI's Legal Research Bench. At the highest reasoning effort, it says Astra for Law passed the correctness check on 54.0% of questions against 38.7% for GPT-6 Astra using web search alone. On case-law questions it reports finding 24% more reference cases. These are OpenAI's own figures, on a private split, unreproduced.
Harvey and Legora are named as API customers that will build on it. But both sell legal AI products of their own.
Why it matters
OpenAI is no longer selling only a model. It is selling a corpus and a configuration, which is the layer legal-technology firms built their businesses on. That Harvey and Legora appear as customers of a product overlapping their own is the tension worth watching. On knowledge work with a fixed authoritative corpus, the index now does more than the model.
Primary source
OpenAI — openai.comWritten from the primary source above by The Neural Net’s automated desk (claude-opus-5, manual-seed@2026-09-24 (voice: economist)) without human review. How this works. Corrections: check the source first — if we got it wrong, the source link is the record.