OpenAI unveiled Astra for Law on September 17, the first vertical edition of its flagship GPT-6 Astra model. It isn't a new model — it's GPT-6 Astra wrapped in a legal research index and a set of specialized instructions — and it competes directly with Harvey, one of the largest legal AI startups on the market, which is also a customer of OpenAI's own API.
Astra for Law doesn't retrain or modify the base model: it takes GPT-6 Astra and pairs it with a legal research index covering U.S. case law, statutes, regulations, court rules, and administrative decisions, drawing on more than 230 million URLs and sources updated daily. On top of that index, OpenAI added 26 new ecosystem plugins connecting ChatGPT to tools firms already use daily, like Relativity (document review) and Clio (case management).
According to OpenAI's own internal evaluation, at maximum reasoning effort Astra for Law passed the general correctness check on 54% of evaluated questions, versus 38.7% for GPT-6 Astra using web search alone — a 40% relative improvement. On questions focused specifically on case law, Astra for Law found 24% more relevant reference cases than the base model with web search.
The most interesting part of the announcement isn't technical, it's strategic: Harvey, one of the most well-funded legal AI products on the market, is built on OpenAI models and is now simultaneously a customer of the Astra for Law API and a direct competitor of the product OpenAI offers to law firms. It's the same pattern OpenAI has already shown with earlier vertical products: sell the infrastructure while, at the same time, launching a first-party application that competes with the companies building on top of that infrastructure.
For companies and law firms in Guatemala and Latin America, Astra for Law is today a product centered on U.S. law — its index of case law, statutes, and court rules covers U.S. sources, not Guatemalan or regional legislation. But the trend still matters: it confirms that frontier labs are packaging generic models into vertical products with proprietary indexes and domain-specific evaluations, rather than leaving each firm to build its own legal RAG from scratch. Firms and corporate legal departments in the region should ask whether their LLM provider has or plans a similar vertical product for local jurisdiction, and carefully vet any accuracy figures they receive from a vendor — always requesting benchmarks run on their own document corpus, not just the figures published by the model maker.
Carlos Montiel is an enterprise AI solutions architect. He implements LLMs, Agents, RAG and orchestrators for companies across Guatemala and Latin America. Reach out for a consultation.
Contact Carlos Montiel