What OpenAI Actually Shipped

OpenAI released Astra for Law on September 17, 2026, a configuration of its GPT-6 Astra model paired with a legal search index spanning more than 230 million URLs of US case law, statutes, regulations, court rules, and administrative decisions, with new sources added daily.

The case law itself comes from the Free Law Project's CourtListener database, a nonprofit open legal archive, rather than a proprietary dataset OpenAI built alone. On Vals AI's Legal Research Bench, a private validation set of 200 questions, Astra for Law answered 54% correctly, compared with 38.7% for GPT-6 Astra using ordinary web search, a meaningful accuracy gain for the specific task of legal research.

The Benchmark Numbers, Side by Side

Four figures capture what changed and by how much, from the independent benchmark score to the size of the underlying index and one customer's own measured result.

MetricResult
Legal Research Bench score, Astra for Law54% of 200 questions correct
Legal Research Bench score, GPT-6 Astra plus web search38.7% of 200 questions correct
Legal search index size230 million-plus URLs, updated daily
Legora's own workflow test upliftabout 40% on a financial-statement review task

The 54% correctness rate is well short of perfect, a detail worth stating plainly rather than letting the accuracy jump alone carry the story: this is a stronger research assistant, not an infallible one.

Two Ways In: Trusted Access and the API

Selected law firms reach Astra for Law directly through a Trusted Access program inside ChatGPT and Codex, where it appears in the model picker as GPT-6 Astra Law, no separate legal-AI vendor required.

An API version, gpt-6-astra-law, is coming later with no confirmed date or pricing. Twenty-six plugins from partner companies, including Thomson Reuters, Harvey, Legora, and iManage, are launching alongside it, meaning the same underlying engine now reaches law firms through at least two separate doors: directly from OpenAI, and wrapped inside the products those partner companies sell.

The Squeeze on Harvey and Legora

Harvey and Legora, two of the better-known legal-AI vendors law firms currently pay to use, are named API customers building their own products on top of Astra for Law. Legora, a Swedish legal-AI company serving firms across Europe, reported roughly a 40% improvement on a financial-statement review task using the new model, completing a tie-out across 41 documents in a single run and catching all four errors it had deliberately planted in a test set, including a hidden gap in a revenue line.

That result is genuinely strong. It also sits next to an uncomfortable fact for Harvey and Legora: the same GPT-6 Astra engine powering their product is available to their own customers directly through OpenAI's Trusted Access program. A firm deciding today between building on a legal-AI vendor or going straight to OpenAI is not just comparing which is smarter right now, it is betting on whether that vendor still exists as a distinct layer once the model maker it depends on can reach the same customer without them.

The Gap Nobody in the Announcement Mentions

CourtListener's case law is a US archive. Astra for Law's index of statutes, court rules, and administrative decisions is built around the US legal system, not the law of Germany, France, or any other EU jurisdiction.

A European firm adopting Astra for Law today gains a genuinely stronger research tool for US-facing matters, cross-border litigation, or American co-counsel work, but not a system that has read its own country's statute book. That distinction matters when a partner is deciding what to tell associates the tool is actually good for, and overselling it as a general legal research assistant for EU law would be a mistake the benchmark numbers do not cover.

Servola Journal

We do this for everyone trying to keep up with what technology is doing to our lives. The people who build it, and the people it happens to. The Servola Journal exists so that what we learn belongs to all of them.

Nobody pays us for this. No ads, no paywall, free to everyone. We just believe that understanding what's happening to all of us shouldn't depend on who can afford to pay for it.

If it gave you something today, tell us to keep going. Follow us, leave a like, or write a positive comment. We read every one, and they are what keeps us going.