CONFIRMED
OpenAI updated Astra for Law on November 17 and now lets every ChatGPT Enterprise customer request access to the legal configuration, including through an application programming interface, or API. The change moves the product beyond its initial circle of invited law firms, but does not make it automatically available to everyone. Access remains subject to the Trusted Access Program. Its proposition is more accurate legal research and preparation, not eliminating a lawyer’s review.
A specialist built on Astra
The product combines GPT-6.2 Astra with a legal search index and specific instructions for investigating questions, weighing authorities and writing answers. It is not simply the general model with a different name. OpenAI says the index has grown 34 percent since September to more than 310 million addresses, adding sources from the United Kingdom, Australia and Canada. For teams moving between research, reading and initial drafting, the proposition is to bring those stages into a single environment.
A lead that still leaves work for lawyers
Artificial Analysis confirms that Astra for Law 6.2 leads its implementation of Harvey’s legal test, LAB-AA v1.1, scoring 20.8 percent against 15 percent for GPT-6.2 Astra at maximum effort. The evaluation includes 120 assignments across 24 legal practice areas and requires every deliverable to meet all criteria without a material invented claim. A panel of model judges grades the work. The result measures performance on this dataset and execution environment, not a universal accuracy rate in actual cases.
Fewer hallucinations do not make answers infallible
The lead matters precisely because absolute performance remains low under the strictest standard. The system averaged 0.03 material hallucinations per assignment in the independent evaluation. That does not justify describing any given response as error free. On Vals AI’s legal research evaluation, OpenAI reports full credit on 53.8 percent of answers. These are different tests and should not be combined or compared as though they measured the same assignment.
The implications for firms and smaller businesses
The practical opportunity is preparation: locating authorities, challenging a contract interpretation and organizing questions for legal counsel. A business can arrive at a consultation with more focused issues, while the lawyer remains responsible for checking sources, currency and application to the case. The announcement does not demonstrate performance on Brazilian law. International expansion of the index alone does not establish adequate coverage of Brazilian legislation or case law, an important distinction for readers operating there.
Broader access still leaves pricing questions
Enterprise customers can request enablement through the admin console; existing authorized users should select gpt-6.2-astra-law. API use also requires a request. The announcement provides no configuration-specific price table, so generating less text in an evaluation does not establish guaranteed invoice savings. For a small practice, subscribing solely because of this launch without clarifying approval, jurisdictional coverage and billing would mean committing before the necessary purchasing information is available.
MaxAssistant’s assessment
The competitive advance is the combination of a model, specialized sources and task-specific instructions, not the replacement of professional responsibility. Broader access makes business evaluation possible; the results explain why that evaluation should start with authorized cases and human review. The strongest purchasing criterion will be time saved after verification, rather than a leaderboard position alone.
Sources
OpenAI: Astra for Law leads in legal intelligence; Artificial Analysis: Harvey LAB-AA v1.1 Benchmark
