OpenAI Launches Astra for Law, a Legal Research Tool Built on GPT-6 Astra

OpenAI has introduced Astra for Law, a legal-focused configuration of its GPT-6 Astra model designed for legal research, analysis, and writing tasks. The product pairs the base model with a dedicated legal search index and task-specific instructions rather than shipping as a separate trained model.

What's in the index

The search index covers US case law, statutes, and regulations across more than 230 million URLs. OpenAI sources this data from the Free Law Project, a nonprofit legal data provider that claims coverage of over 99.9 percent of published US legal precedents.

Benchmark results

OpenAI ran its own evaluation using Vals AI's Legal Research Bench, a 200-question test. According to OpenAI, Astra for Law passed 54 percent of questions, compared to 38.7 percent for the standard GPT-6 Astra model using plain web search. These figures come from OpenAI's internal testing and have not been independently verified.

Access and integrations

OpenAI says API customers, including legal tech companies Harvey and Legora, can build products on top of Astra for Law. The company is also rolling out a Trusted Access program for law firms that includes privacy controls such as zero data retention — meaning firm data isn't stored or used for training.

Alongside the launch, OpenAI is releasing 26 plugins connecting Astra for Law to existing legal software, including Relativity and Clio, two platforms widely used in e-discovery and law firm case management.

Competitive context

OpenAI isn't the only major AI lab targeting legal work. Anthropic has also been expanding its footprint in the legal market, positioning Claude models for similar research and drafting tasks. Legal research and document review are among the most cited early use cases for AI in professional services, given the high cost of associate and paralegal time and the structured, text-heavy nature of legal work.

What this means

Astra for Law is not a new model — it's GPT-6 Astra wrapped with a legal-specific retrieval layer and prompting. That distinction matters: the performance gain from 38.7 percent to 54 percent on the benchmark comes primarily from better grounding in authoritative case law data, not from any change to the underlying model's reasoning capability. This is a pattern worth watching across the industry — vertical AI products increasingly differentiate through domain-specific retrieval and integrations rather than through frontier model improvements alone.

For law firms, the appeal is straightforward: a searchable index with near-complete coverage of US case law reduces the risk of hallucinated citations, a well-documented failure mode that has already led to sanctions against lawyers using general-purpose chatbots for legal research. Zero data retention and the Trusted Access program also address a major adoption barrier — firms' reluctance to send confidential client data to third-party AI services.

The real test will be adoption by legal tech incumbents. Harvey and Legora building on OpenAI's API suggests OpenAI is positioning itself as infrastructure for the legal AI stack rather than competing directly with these vertical specialists — a strategy likely to determine whether Astra for Law becomes a meaningful revenue line or a checkbox feature.