OpenAI expands Astra for Law to all ChatGPT Enterprise customers
OpenAI updated its legal-specific AI configuration with a newer base model and a larger international search index.
The short version
- Astra for Law is now available to all ChatGPT Enterprise customers on a request basis.
- The tool was updated to use the GPT-6.2 Astra base model, improving reasoning and performance.
- The legal search index grew by 34% to cover over 310 million URLs, including data from the U.K., Australia, and Canada.
- OpenAI claims the model achieved the top score on a specific benchmark of legal agent work developed by Harvey.
Why it matters
This move signals a shift from experimental pilots to a standardized tool for large legal firms. It provides enterprise users with a model specifically tuned to handle complex legal research and drafting with lower hallucination rates.
Expanding access to legal AI
OpenAI has officially opened access to Astra for Law for every ChatGPT Enterprise customer. While the tool launched in September as a limited pilot for a small group of "Trusted Access" firms, it is now available to all enterprise users on a request basis. This expansion allows a wider range of legal organizations to utilize a model specifically configured for legal work.
The update moves the tool onto the newer GPT-6.2 Astra base model. This transition is part of OpenAI's broader strategy to integrate frontier models into specialized professional fields. By providing a dedicated configuration, OpenAI aims to offer more reliable results than a general-purpose AI might provide for complex legal tasks.
Users can now request access to the tool through the Trusted Access program in their admin console. Additionally, OpenAI is offering API access for organizations that want to build Astra for Law into their own custom software. This dual approach caters to both ready-made users and developers who need to integrate the technology into existing workflows.
Technical updates and performance
The new version of Astra for Law includes specific instructions for researching questions, weighing authorities, and writing answers. These instructions are layered on top of the GPT-6.2 Astra base model to ensure the AI follows legal logic. This configuration approach was emphasized by Jason Boehmig, who noted that the tool is not a new model but a specialized configuration of an existing one.
Performance gains were measured using the Artificial Analysis version of Harvey’s Legal Agent Benchmark (LAB-AA). This benchmark tests models on 120 assignments across 24 legal areas, such as reviewing commercial leases or drafting deposition outlines. Astra for Law earned full credit on 20.8% of these assignments, outperforming the base GPT-6.2 Astra model, which scored 15.0%.
The model also showed significant improvements in accuracy and efficiency. It reached its benchmark scores with less than half the output of the highest-scoring model from another developer, using about 64,000 tokens per assignment. Furthermore, it tied for the lowest rate of material hallucinations in the comparison, averaging only 0.03 per assignment.
A larger global search index
A critical component of Astra for Law is its legal search index, which provides the data the model uses to find relevant laws. Since its initial launch, OpenAI has grown this index by 34%, adding more than 80 million new URLs. The index now covers more than 310 million URLs, including material from the U.K., Australia, and Canada.
The updated index now spans more than 99.9% of published U.S. precedential case law. It draws on a variety of sources, including statutes, regulations, court rules, and administrative decisions. This expansion ensures that the model can reference a broader range of international and domestic legal authorities when generating responses.
By enlarging the index, OpenAI is attempting to make the tool more useful for international firms. The inclusion of non-U.S. data allows for more comprehensive research across different jurisdictions. This growth follows the initial launch which focused primarily on U.S. case law from the CourtListener database.
Benchmark results and feedback
OpenAI reported that Astra for Law earned full credit on 112 of 208 questions on the Vals AI Legal Research Bench. This represents a 53.8% success rate, an improvement from the 46% success rate reported in September. While it is considered a "top performer" on this bench, it is not currently the leading model in that specific category.
The improvements in the 6.2 version also resulted in better-reasoned outputs. Julio Pereyra, head of legal research at Harvey, stated that the model writes briefs that are preferred to those of the base model nearly 90% of the time. These improvements in depth were achieved without regressing on the model's accuracy or increasing its hallucination rate.
The data suggests that while the new base model provided the majority of the performance jump, the legal configuration added a measurable boost. The configuration provided an additional 1.6 percentage points of success on Harvey's original LAB. This confirms that the specific legal instructions and tools are providing a distinct advantage over using a general model.
The road ahead for legal AI
This release is part of a larger trend where major AI labs are creating specialized versions of their models for the legal industry. Anthropic, Google, and Microsoft have all launched similar legal-focused products within the last year. OpenAI's move to open Astra for Law to all enterprise customers marks a significant step toward standardizing these tools in the workplace.
While this update focused heavily on model performance and index size, it did not introduce new plugins or custom skills. These features were part of the initial September launch. OpenAI has stated that they are only at the beginning of their investment in the legal field and that more developments are coming in the future.
Future updates may address broader concerns like access to justice, which were mentioned as a paramount concern during the initial launch. For now, the focus remains on refining the accuracy and depth of the model. The company intends to continue developing the tool to see how the legal profession adopts and utilizes the technology.
What to watch
- Whether OpenAI will release new community plugins or custom skills for Astra for Law.
- How other legal AI providers like Harvey and Anthropic respond to OpenAI's wider enterprise access.
- Future updates to the search index to include more international jurisdictions.
Try it
If you have a ChatGPT Enterprise account, check your admin console to see if you can request access to the Astra for Law configuration.
Sources and links
- Officialreleased an updated version of Astra for Law openai.com
- ReportingLawSites | by Robert Ambrogi: the article we sourced
First reported by LawSites | by Robert Ambrogi. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.