OpenAI’s legal-focused GPT-6 Astra configuration brings specialised search, legal reasoning and workflow tools to law firms, intensifying competition with Anthropic and Thomson Reuters in a market where accuracy, citations, privacy and professional reliability are critical

Artificial intelligence is moving deeper into professional services, and the legal industry is emerging as one of the most competitive battlegrounds for specialised AI. OpenAI has expanded its enterprise strategy with Astra for Law, a legal-focused configuration of its GPT-6 Astra model designed for research, analysis, drafting and professional legal workflows.

The launch comes as legal technology companies and established information providers develop their own AI systems. Anthropic is considering a new model amid competitive pressure from OpenAI, while Thomson Reuters has already developed its proprietary Thomson large language model for legal and compliance applications.

The shift reflects a broader change in enterprise AI. Instead of relying solely on general-purpose chatbots, professional users increasingly want systems that combine powerful models with authoritative databases, specialised workflows, traceable citations and controls designed for sensitive client information.

What is Astra for Law?

Astra for Law is built on GPT-6 Astra and combines the model with a dedicated legal search index, specialised instructions for legal analysis and writing, and tools designed around professional legal workflows.

OpenAI says the system can search US case law, statutes, regulations, court rules and administrative decisions through an index covering more than 230 million URLs, with new sources added daily. It also incorporates case-law material from the Free Law Project's CourtListener collection.

The service is initially being made available to selected US law firms through OpenAI's Trusted Access programme, with access through ChatGPT and Codex. API availability is planned for developers and legal technology companies.

That approach positions Astra for Law not simply as another chatbot, but as an AI foundation that legal software providers can integrate into their own products and workflows.

Legal search is at the centre of the offering

One of the biggest challenges in legal AI is not simply generating fluent answers. Lawyers need to locate the right authority, understand how a precedent applies and verify that a citation actually supports the proposition being made.

Astra for Law's dedicated legal search infrastructure is intended to address that problem.

OpenAI says the system can search across cases, statutes, regulations and administrative decisions and can help identify authorities that weaken an argument as well as those that support it. The model can then apply the research to client facts, arguments and deal terms.

This is significant because legal work often requires finding counterarguments rather than merely generating a persuasive first answer.

OpenAI reports higher performance on its legal benchmark

OpenAI evaluated Astra for Law using 200 US legal research questions from the private validation set of Vals AI's Legal Research Bench.

At the highest reasoning setting, OpenAI reported that Astra for Law passed the benchmark's overall correctness check on 54 per cent of questions, compared with 38.7 per cent for GPT-6 Astra using web search alone. OpenAI characterised the difference as a 40 per cent relative improvement.

The company also reported that Astra for Law found 24 per cent more reference cases on case-law questions and retrieved up to 54 per cent more relevant passages from correct court opinions on an audited set.

These figures are OpenAI's own evaluation results rather than an independent industry-wide assessment. The benchmark therefore provides an indication of the company's reported performance, but lawyers and legal departments will still need to evaluate accuracy, citation quality and reliability in their own workflows.

Privacy and governance become critical

Legal AI operates in a particularly sensitive environment because firms routinely handle confidential client information, litigation strategies, contracts and commercially sensitive transactions.

OpenAI says Astra for Law includes privacy and governance controls designed for confidential client work. The company's legal offering also connects with specialist tools through ecosystem plugins.

The importance of these controls goes beyond technical performance. Law firms adopting AI must consider confidentiality, data handling, professional responsibility, client expectations and the need for lawyers to verify AI-generated work.

The industry has therefore increasingly shifted the conversation from whether AI can produce an answer to whether the answer can be trusted and audited.

Anthropic faces increasing competitive pressure

The Astra for Law launch comes shortly after OpenAI's broader GPT-6 Astra release, which has reportedly gained traction among enterprise users.

Reuters reported that Anthropic is considering launching a new AI model to respond to OpenAI's momentum. The company has not made a final decision on the release, according to people familiar with the matter cited by Reuters.

The competitive pressure is particularly notable because Anthropic has established a strong position in enterprise AI.

Reuters reported that OpenAI's Astra accounted for around 13 per cent of enterprise AI spending tracked by corporate expense platform Ramp, compared with about 8 per cent for Anthropic's Claude Fable. Reuters also reported that OpenAI models led Anthropic's models in spending on OpenRouter during the latest week measured.

At the same time, Anthropic continues to have a larger reported annualised revenue run rate, according to Reuters, illustrating that market share indicators and overall revenue leadership are not necessarily moving in the same direction.

Thomson Reuters takes a different route

The legal AI race is not limited to frontier-model companies.

Thomson Reuters has developed Thomson, its own proprietary large language model specifically designed for high-stakes professional work, including legal and compliance applications.

The company says Thomson was trained using proprietary content from Westlaw, Practical Law, Checkpoint and Reuters, combined with validation from subject-matter experts.

This gives Thomson Reuters a different competitive advantage from general-purpose AI companies.

Rather than primarily competing on model capability, the company can combine its model with decades of proprietary legal information, established professional workflows and relationships with law firms and legal departments.

CoCounsel provides Thomson with a distribution advantage

Thomson Reuters is already deploying Thomson inside CoCounsel Legal.

Its Tabular Analysis tool can review up to 10,000 documents against as many as 100 questions, with results designed to remain traceable to their sources.

The company says more than one million professionals use CoCounsel, giving Thomson Reuters an existing user base through which its proprietary AI can be distributed.

That installed base could become strategically important as legal AI adoption moves from experimentation into regular professional workflows.

The competition is shifting from models to ecosystems

The legal AI market increasingly appears to be developing around three interconnected layers.

The first is the foundation model — the underlying AI capable of reasoning, understanding documents and generating text.

The second is the knowledge layer — authoritative case law, statutes, regulations, legal commentary and proprietary professional information.

The third is the workflow layer — tools that connect AI to document management, contract review, litigation discovery, research, drafting and other daily legal processes.

OpenAI's strategy emphasises the first layer while increasingly strengthening the second and third through legal search and ecosystem integrations.

Thomson Reuters has a structural advantage in the knowledge and workflow layers through Westlaw, Practical Law and CoCounsel.

Anthropic, meanwhile, is competing primarily through its frontier models and enterprise adoption while continuing to develop its broader professional AI strategy.

Why accuracy matters more in legal AI

The legal industry has little tolerance for confident but incorrect AI output.

A fabricated case, incorrect citation or misunderstood contractual provision can create financial, regulatory or litigation risks.

That is why legal AI providers increasingly emphasise citation grounding, source verification, human review and auditability.

Thomson Reuters, for example, says its CoCounsel ecosystem is designed around authoritative sources and traceable outputs. Its Deep Research Verify feature is intended to check whether cited Westlaw and Practical Law authority supports specific legal assertions.

OpenAI's legal search index is similarly aimed at improving the connection between generated analysis and underlying legal authorities.

Implications for legal technology companies

The emergence of Astra for Law could create both opportunities and competitive pressure for legal technology companies.

Companies such as Harvey and Legora can use OpenAI's underlying technology to build specialised applications rather than developing every component of the AI stack themselves. OpenAI has specifically identified these companies as API customers able to build on Astra for Law.

This could accelerate innovation in areas such as litigation research, contract analysis, due diligence, compliance and transaction workflows.

At the same time, legal software providers will need to differentiate themselves through proprietary data, workflow integration, customer relationships, security and domain expertise rather than simply access to a powerful underlying model.

What it means for law firms

For law firms, the increasing competition among AI providers could expand the range of tools available for routine and research-intensive work.

Potential use cases include legal research, document analysis, contract review, due diligence, drafting, identifying counterarguments and extracting structured information from large document collections.

However, the technology does not remove the need for professional judgment.

Legal professionals remain responsible for checking authorities, interpreting facts, assessing risks and ensuring that final work complies with professional and client requirements.

The broader enterprise AI race

The legal sector is becoming an important test case for the broader enterprise AI market.

Professional users generally need more than raw model intelligence. They need specialised data, security, workflow integration, predictable performance and clear accountability.

This helps explain why companies across the technology and information-services industries are building vertical AI products.

The same pattern is emerging in financial services, healthcare, cybersecurity and other sectors where specialised knowledge and accuracy are critical.

Market Outlook

OpenAI's Astra for Law marks another step in the shift from general-purpose AI towards specialised professional systems. Its combination of GPT-6 Astra, legal search, professional instructions and workflow integrations places greater emphasis on the complete legal-AI ecosystem rather than the underlying model alone.

The competitive landscape is becoming more complex. OpenAI is expanding its frontier-model capabilities into professional applications, Anthropic is evaluating its response to Astra's reported enterprise momentum, and Thomson Reuters is building a proprietary model around its established legal information and workflow ecosystem.

For the legal industry, the critical metric will ultimately be the reliability and usefulness of AI in real professional workflows rather than benchmark scores alone. Citation accuracy, confidentiality, integration, auditability and human oversight are likely to remain central considerations.

The development also signals a broader trend for enterprise technology: as AI models become more capable, competition is increasingly moving beyond raw intelligence towards specialised data, trusted information, domain expertise and workflow integration.