top of page

Astra for Law 6.2: What OpenAI's New Legal AI Means for Small Law Firms

4 hours ago
3 min read

OpenAI announced Astra for Law 6.2 on November 17, an upgrade to the legal research version of its newest AI model. If you run a law firm, expect to hear about it from vendors, from clients, and from your own associates. Here is what was announced, what the numbers really say, and what we would do about it this month.


What OpenAI announced

Astra for Law is not a separate product you install. It is OpenAI’s GPT-6.2 Astra model paired with a search index of legal sources and a set of instructions for researching a question, weighing the authorities and writing up the answer.


The headline change is the index. OpenAI says it has grown by about a third since September, now covers more than 99.9% of published U.S. precedential case law, and has started adding sources from the UK, Australia and Canada. Cooley, Ropes & Gray and Sullivan & Cromwell have been using early versions since September.



What the numbers say

OpenAI published results on benchmarks built with Harvey, Vals AI and others. On Vals AI’s Legal Research Bench, Astra for Law earned full credit on 53.8% of questions, up from about 46% in September. On Harvey’s assignment benchmark it ranked first, with full credit on 20.8% of assignments against 15.0% for the standard model.


On invented citations and facts, OpenAI reports an average of 0.03 material errors per assignment, tied for the lowest in its comparison.



Read those numbers both ways

The improvement is real. So is the gap. Full credit on 53.8% of research questions means close to half of the answers still missed something a grader was looking for. These are also results a vendor chose to publish about its own product.


A low error rate is not zero. The attorney who signs the filing still owns every citation in it, whichever tool found it.


Can a small firm get it today?

Not directly. For now, access goes through OpenAI’s Trusted Access Program: ChatGPT Enterprise customers can request it, and developers can apply for API access. OpenAI has not published pricing. Most small and mid-sized firms will first meet this technology inside the research and practice tools they already pay for, as those vendors build on it.


Five things to do this month

You do not need this model to have an AI risk, or an AI opportunity. Your people are very likely using some AI tool already. These five steps apply whichever one it is.



  1. Find out what is already in use. Ask every attorney and staff member which AI tools they use for work, including free ones on personal accounts.

  2. Write a one-page AI policy. List the approved tools and state plainly what may never be pasted into them, starting with confidential client information.

  3. Make citation checking a rule. Every case, statute and quote an AI tool produces gets pulled and read by a person before it leaves the firm.

  4. Read the data terms. For each approved tool, confirm whether your inputs are used to train the model, where they are stored and who can see them.

  5. Lock down the accounts. Use firm-managed accounts with multi-factor authentication, so access ends when someone leaves.


Where Edward Technology fits

We are an IT company, not a law firm, and nothing here is legal or ethics advice. Your state bar’s guidance on AI comes first. What we handle is the technical side: finding which AI tools are in use on your network, setting up firm-managed accounts and Microsoft 365 controls, and turning your policy into settings that enforce it.



Source: OpenAI, Astra for Law 6.2 announcement, November 17, 2026.

Comments


Barrington Ice House Mall

200 Applebee St.

Barrington, IL 60010

Elmhurst City Centre

120 N. York St.

Elmhurst, IL 60126

  • YouTube Social  Icon
  • LinkedIn Social Icon

YouTube

LinkedIn

bottom of page