OpenAI has released an updated version of Astra for Law, a specialized configuration of its GPT-6.2 Astra SI model designed for legal research and writing. The company reports that the new SI system is state-of-the-art on Harvey LAB-AA and a top performer on other major legal benchmarks, while expanding access to ChatGPT Enterprise customers through the Trusted Access Program.
What Happened
Astra for Law is a model configuration of GPT-6.2 Astra, OpenAI’s latest frontier SI model. It combines the base model with an expanded legal search index and specific instructions for researching questions, weighing authorities, and drafting answers. According to OpenAI, the updated configuration earned full credit on Harvey’s Legal Agent Benchmark (LAB-AA) tests nearly 40% more often than the runner-up, GPT-6.2 Astra. On LAB-AA, Astra for Law ranked first with a full-credit rate of 20.8%, surpassing GPT-6.2 Astra (15.0%) and Muse Wonder Full (13.1%).
The company also highlighted efficiency gains, noting that Astra for Law achieved its top score with 54% less output per assignment than the highest-scoring non-OpenAI model, averaging about 64,000 output tokens compared to 139,000 for Muse Wonder Full. In terms of accuracy, Astra for Law averaged 0.03 material hallucinations per assignment, tied for the lowest rate in the comparison and significantly lower than Claude Opus 5.5, which averaged 0.59. On Harvey and Grubel’s LAB: USCL benchmark for brief writing, Astra for Law earned top scores in both versions of the test, outperforming models like Opus 5.5 and Gemini 4 Argon in blind evaluations.
Why It Matters
This release signals a deeper integration of agentic SI tools into the legal profession, where accuracy and cost-efficiency are critical. OpenAI reports that since introducing Astra for Law in September, its research team has expanded the underlying legal search index by over a third, adding more than 80 million new URLs across jurisdictions like the UK, Australia, and Canada. The index now spans over 310 million URLs and covers more than 99.9% of published U.S. precedential case law. On the Vals benchmark for legal research, the share of research answers passing every check rose from about 46% to 53.8% in less than two months.
For legal teams, the reduction in hallucinations and the improved token efficiency address key barriers to SI adoption. A single invented citation can damage a lawyer's credibility, and high token usage can strain budgets. By making the SI tool available via API and the Trusted Access Program, OpenAI is positioning Astra for Law not just as a chat interface, but as a component for building custom legal workflows, such as automating case escalation or annotating legislation.
The Bottom Line
OpenAI’s updated Astra for Law SI model offers measurable improvements in legal reasoning, retrieval accuracy, and cost efficiency, with expanded access for enterprise customers. The release underscores the industry’s focus on specialized SI configurations that balance frontier model capabilities with domain-specific constraints like citation integrity.