OpenAI releases updated Astra for Law, leading legal benchmarks with 20.8% full-credit rate
·2026.11.17 09:00
Key point
The updated model configuration achieves 53.8% full credit on Vals AI's Legal Research Bench and is now available to all ChatGPT Enterprise customers.
Details
- OpenAI has released an updated Astra for Law, a model configuration of GPT-6.2 Astra, which ranks first on Harvey's LAB-AA benchmark with a 20.8% full-credit rate, outperforming GPT-6.2 Astra (15.0%) and Muse Wonder Full (13.1%).
- The configuration demonstrates significant efficiency and accuracy, using 54% fewer output tokens per assignment than the top non-OpenAI model and recording 0.03 material hallucinations per assignment, a 95% reduction compared to Claude Opus 5.5.
- Legal research performance improved rapidly since September, with the share of answers passing all checks on Vals AI's Legal Research Bench rising from 46% to 53.8% in under two months, while the legal search index expanded by 34% to over 310 million URLs.
- Access to the updated model is now available to all ChatGPT Enterprise customers through the Trusted Access Program, including via API, following initial pilots with firms such as Cooley, Ropes & Gray, and Sullivan & Cromwell.
Benchmark Performance
- On Harvey's LAB-AA, Astra for Law earned full credit on 20.8% of 120 assignments, a 59% improvement over the highest-scoring non-OpenAI model.
- On LAB: USCL for brief writing, the model scored 1,844 in the research-and-draft version, compared to 1,817 for Opus 5.5, with Harvey's judges preferring its briefs nearly 90% of the time over GPT-6 Astra.
- On Harvey's own Legal Agent Benchmark, the legal configuration raised the full-credit rate to 16.2%, an 11% increase over the base GPT-6.2 Astra model.
Availability and Features
- The model pairs GPT-6.2 Astra with an expanded legal search index covering over 99.9% of published U.S. precedential case law and jurisdictions including the UK, Australia, and Canada.
- Early users are employing the tool for argument sparring, distinguishing legal requirements from recommendations, and mapping regulatory rules for new business activities.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.