OpenAI's Astra for Law targets Thomson Reuters and the legal-tech incumbents

OpenAI is putting its most powerful model to work on a market long dominated by Thomson Reuters and LexisNexis. On 17 September 2026 the company launched Astra for Law, a version of GPT-6 Astra configured for legal research and drafting, appearing in the model picker as "GPT-6 Astra Law". It is going first to a short list of elite US firms, not the broader market.

The product is built around a Legal Search Index of more than 230 million URLs covering US case law, statutes, regulations, court rules and administrative decisions, updated daily. Case law comes through a partnership with the nonprofit Free Law Project's CourtListener database, which OpenAI says covers more than 99.9% of published US precedential case law. That is a direct challenge to the proprietary databases legal-research incumbents have charged premiums for.

OpenAI paired the model with 26 partner plugins from vendors including Thomson Reuters, Intapp, Relativity, Clio and DeepJudge, plus nine community plugins offering 47 skills. ChatGPT for Word is now generally available. For firms handling privileged material, eligible customers get zero data retention and default exclusion from human review.

On OpenAI's own benchmark, 200 US legal research questions from the Vals Legal Research Bench, GPT-6 Astra Law scored 54.0% correctness against 38.7% for standard GPT-6 Astra with web search, and found 24% more relevant cases on case-law questions. There is no independent comparison yet against incumbent legal research platforms, according to Artificial Lawyer.

Access is limited for now to selected US law firms through Trusted Access and Codex, with an API to follow. Early adopters named by OpenAI include Latham & Watkins, Ropes & Gray, Cooley and Sullivan & Cromwell, and legal-AI platforms Harvey and Legora are building it into their own products rather than treating it as a rival. Ropes & Gray is already using it inside an M&A due-diligence system that traces findings back to source documents and flags whether customer contracts require notice or consent on an acquisition.

By signing Thomson Reuters as a plugin partner rather than bypassing it, OpenAI is buying distribution instead of picking a fight with the incumbent it most threatens. That hedges the risk that law firms, which move slowly and care about liability, balk at an unproven model for high-stakes research. The zero-retention privacy terms are aimed squarely at that caution.

The real test is the elite-firm rollout. Latham & Watkins, Ropes & Gray, Cooley and Sullivan & Cromwell carry weight in the profession; if their use cases prove out, Astra for Law becomes a reference sale for every mid-size firm watching from the sidelines. Harvey and Legora building on top of it, rather than competing head-on, suggests the legal-AI layer is consolidating around OpenAI's model rather than fragmenting further.