Our most powerful model, configured into a new AI foundation for law.
Today, we’re introducing Astra for Law: a new foundation for law firms and legal technology companies to build AI products and workflows around their expertise. It combines GPT‑6 Astra, our latest and most powerful model, with settings, tools, and context tailored for professional legal work.
API customers including Harvey and Legora will be able to build on Astra for Law, bringing this intelligence into their own products and workflows. As our frontier models advance, we’ll bring these legal capabilities to our latest models.
We are also expanding our work on privacy and governance to give law firms specific controls for confidential client work. Firms can also customize Astra for Law using our 26 new ecosystem plugins that connect ChatGPT to the specialist tools firms already use, like Relativity and Clio.
Frontier intelligence for law
Astra for Law combines GPT‑6 Astra with a powerful legal search index and instructions for legal analysis and writing. Together, they amplify Astra’s capabilities across the legal practice, while giving firms and legal technology companies the freedom to build their own applications and workflows.
Legal research: from facts to a supported answer
Our new legal search index is one of the tools Astra for Law can use. Legal research often begins with finding the exact right authority, locating the relevant passages, and understanding how relevant and binding they are to the situation at hand. The index helps Astra for Law do that work, and complements the licensed content and specialist products firms rely on from providers such as Thomson Reuters.
By using the legal search index, Astra for Law can search U.S. case law, statutes, regulations, court rules, and administrative decisions across a corpus of more than 230 million URLs, with sources added daily. Our work with Free Law Project, the nonprofit behind CourtListener, brings its case-law collection covering more than 99.9% of published U.S. precedential case law into this research experience.
To measure how this configuration improves legal research, we tested Astra for Law’s complete setup on 200 U.S. legal research questions from the private validation set of Vals AI’s Legal Research Bench. This benchmark measures how well the model can find relevant sources and passages, and how well its research answers meet the evaluation criteria.
At the highest reasoning effort for both systems, Astra for Law passed the evaluation’s overall correctness check on 54.0% of questions, compared with 38.7% for GPT‑6 Astra using web search alone – a 40% relative improvement. Astra for Law also produces more comprehensive answers.
On case-law-focused questions, Astra for Law found 24% more reference cases than GPT‑6 Astra using web search alone at the highest reasoning effort. On the audited set of target passages, it retrieved up to 54% more relevant passages from the correct court opinions, when comparing the systems at the same reasoning effort.
The result is a stronger research foundation for advising on a deal, assessing a dispute, or developing a legal strategy, with reliable authorities the lawyer can examine for herself.
Astra for Law and GPT‑6 Astra’s performance on the Vals AI Legal Research Bench validation set, across reasoning effort settings.
Improving performance on end-to-end legal workflows
Legal research is only the first step. Custom instructions for legal analysis and writing guide Astra for Law in applying that research to the client’s facts, developing arguments or deal terms, and identifying weaknesses and uncertainty. That can mean distinguishing a court’s holding from its other observations, addressing cases that weaken an argument, or explaining how a contract exception shifts risk between the parties.
For example, when prompted to identify good law with similar fact patterns, Astra for Law could both pinpoint relevant precedent and match fact patterns better than other frontier models:
Prompt
Our client signed a five-year agreement to manufacture a retailer’s private-label products after being told the retailer’s comparable suppliers historically averaged about $8 million in annual orders. The contract commits the retailer to $2 million in annual purchases and leaves additional orders to its discretion. Before signing, our client knew the retailer was revising how it allocated orders among suppliers. Orders have barely exceeded the minimum. Internal records suggest the historical average was much lower. The retailer says our client accepted both the minimum and the new process. Find the closest factual precedent and write a memo on our client’s misrepresentation claim.
Claude Fable 5.1
MEMORANDUM — Misrepresentation Claim Against Retailer
Closest precedent. Paragon Metals Holdings v. Smith, C.A. No. N21C-12-090-SKR-CCLD (Del. Super. Aug. 13, 2025). A buyer alleged the seller concealed that two major customers would materially cut volumes. After a bench trial, the court rejected the fraud claim, concluding the buyers failed to conduct sufficient diligence and ignored repeated, if inartful, disclosures of the very information underlying their claim. Notably, the court assumed certain representations may have been false yet still ruled for the sellers because the buyers could not prove justifiable reliance.
Application. Our strongest fact is that the "$8 million average" was a statement of historical fact, not a forecast, and internal records apparently show it was false. That distinguishes the ordinary "predictions aren't actionable" defense. The $2 million minimum does not disclaim the historical figure; it allocates future risk, not past fact.
Our weakest fact mirrors Paragon: the client knew allocation methodology was changing and signed anyway. The retailer will argue that knowledge made reliance on old averages unreasonable and that the minimum captured the bargained-for risk. Anti-reliance or integration clauses in the agreement would compound this.
Recommendation. Plead fraudulent inducement (and negligent misrepresentation alternatively) centered on the false historical average; develop evidence that the process change was presented as neutral to volume. Confirm governing law—Delaware is notably strict on reliance.
Astra for Law
Our client has a viable claim based on false history, not a right to $8 million in annual orders.
The closest analogue is Oliver Wyman, Inc. v. Eielson, 282 F. Supp. 3d 684, 700–03 (S.D.N.Y. 2017), applying Massachusetts law. Consulting-business sellers accepted four-year employment agreements guaranteeing $425,000, with discretionary increases, after representations of approximately $1 million average partner compensation. They knew the compensation system was changing. The court denied summary judgment on fraud and negligent-misrepresentation counterclaims concerning the historical average: evidence disputed its accuracy, and the contractual floor did not contradict past results.
Similarly, the retailer’s approximate $8 million historical average is a verifiable factual assertion. Accepting a $2 million minimum and discretionary additional orders does not necessarily surrender reliance on truthful historical information. But Oliver Wyman emphasized that no evidence showed the revised system would fundamentally change compensation or render historical averages irrelevant. Id. at 701–02. The retailer’s strongest defense is proof our client understood the allocation changes would make past supplier volumes uninformative—not simply that changes were announced.
For fraud, our client must prove knowing material falsity, intent to induce signing, actual reasonable reliance, and resulting loss. Negligent misrepresentation may instead rest on deficient care in supplying business information. Obtain supplier-level records matching the represented period and comparator group, the speaker’s contemporaneous communications, and our client’s decision records.
Given the same prompt, Astra for Law returned two closely matching precedents; in the litigation example, Claude Fable 5.1 returned a holding that had been reversed on appeal, while in the transactional example it reported finding no such case.
Astra for Law will be initially offered to selected law firms through Trusted Access in ChatGPT and Codex, and will be coming soon to the API. It will appear in the model picker as “GPT‑6 Astra Law” and in the API as gpt-6-astra-law.
“We were grateful to preview early versions of Astra for Law, which were built with legal use-cases in mind. Across both litigation and transactional matters, the models demonstrated impressive research depth and sensitivity to authority. Even at this early stage, they felt like a significant step toward legal-focused AI that is carefully grounded in research that is both current and comprehensive.”
Legal-grade trust and controls
Law firms need to protect client confidences and control how AI is used in their practice. We’ve created a special Trusted Access Program for eligible law firms to give lawyers and people working under their supervision access to Astra for Law for professional legal work. For eligible firms, the offering includes Zero Data Retention (ZDR) on our API, and usage of ChatGPT Enterprise is excluded from human review by default.
We are also working with Latham & Watkins, a leader in AI governance, to design for information permissions, ethical walls, client instructions, and firm oversight.
“As AI becomes more capable, so too does the ability to deploy it in environments that demand rigorous governance, oversight, and accountability. This collaboration builds on Latham’s broader investments in AI development, governance, and infrastructure across the firm, which underpin our enterprise-wide strategy for responsible AI.”
Build ChatGPT around your firm’s expertise
With frontier intelligence and the right controls, firms can turn their own precedents, methods, and judgment into AI tools and workflows built to their standards.
Working with selected firms, our forward-deployed engineers have been adapting ChatGPT Enterprise with custom interfaces and integrations to proprietary data, creating tools for each firm’s workflows:
Sullivan & Cromwell built an agreement analyzer that brings the firm’s negotiating playbooks and selected precedents into the review of a new deal. It helps lawyers spot risks that emerge when provisions are read together, then turns those findings into proposed redlines and draft client advice they can challenge and refine.
Ropes & Gray built a deal diligence system around how its lawyers work through a data room and decide what matters to the deal. It helps them trace findings back to the source and pinpoint questions that could affect an acquisition, such as whether key customer contracts require notice or consent.
Cooley built GO Public to bring its capital markets expertise into how companies prepare to go public, from drafting the IPO filing to identifying the risks that deserve management’s attention. When the deal changes, it carries that change across the filing so lawyers can review the implications together.
“Our collaboration with OpenAI has allowed us to rethink how this work gets done – moving lawyers and management teams more quickly through intensive preparation and into questions that require judgment, market experience and strategic thinking.”
Firms can build with their own teams and partner products, with permitted sources and review processes defined for the work.
Frontier intelligence, connected to the most trusted tools in legal
We’re proud to work with the specialist companies who are advancing legal AI. Today, we’re launching 26 partner-built plugins that help firms go deeper with the tools and knowledge they already use. These plugins cover the practice and business of law. With iManage, a lawyer can draft a negotiation brief in ChatGPT and save it to the matter file; Intapp can surface activities that may need a time entry for review; DeepJudge can bring prior deals into a comparison. Thomson Reuters is bringing HighQ matter context into ChatGPT and previewing a forthcoming CoCounsel Legal connector.
“As AI becomes more open and interoperable, the value is not in connectivity alone. Legal professionals need more than access to information. They need trusted intelligence, relevant enterprise and matter context, purpose built legal capabilities, and the governance required for high stakes work. That is what Thomson Reuters delivers through HighQ and CoCounsel Legal. CoCounsel remains the trusted professional AI system designed to help complete that work. Our work with OpenAI helps make these capabilities available in the environments customers choose, while preserving the accuracy, confidentiality, and accountability they depend on.”
The launch includes 9 community plugins from lawyers and legal engineers at LegalQuants, LECG, and Skills.law, with 47 custom skills that practitioners can adapt, extend, or draw inspiration from to do their work in ChatGPT. Community plugins give the people closest to the work a way to keep setting the standard for what these skills can do. We’re also making ChatGPT for Word generally available today, so lawyers can proofread, get suggested edits, and flag formatting issues in the tool they already rely on for drafting.
Our approach is open and composable: firms can use specialist products, bring their own tools and knowledge, and adapt community-built skills to their practice. Partners can keep developing their own applications and workflows, while firms choose how those capabilities fit together. Building on OpenAI should mean getting more from the ecosystem, not replacing it.
Build with us
OpenAI is investing in law for the long term. We’ll keep advancing Astra for Law’s model, settings, tools, and instructions together, guided by rigorous evaluations and feedback from lawyers and legal technology partners. We’ll also keep improving ChatGPT so firms can build around their expertise and connect the tools they use.
Our work with Wachtell, Lipton, Rosen & Katz brings the firm’s litigation and corporate expertise together with our frontier research and engineering to explore how AI can support the work involved in providing sophisticated legal judgment.
“We’ve appreciated the technical depth and willingness to listen that OpenAI’s forward deployed engineering team has brought to our conversations about AI's role in supporting legal work, and are excited about the possibilities of AI to shape the next era of legal practice.”
The most ambitious applications will come from the people who practice law and the companies building alongside them. We’re committed to working with this ecosystem so firms can bring their own expertise, standards, and judgment to AI—and serve clients in ways only they can.
To explore early access to Astra for Law, build it into your product, or develop tools around your firm’s expertise, contact OpenAI.