The demand for artificial intelligence (AI) systems tailored to legal work has grown as law firms and legal technology companies look for tools that can handle research and other professional tasks. In his 2023 year-end report, US Chief Justice John Roberts warned about the risks of generative AI producing inaccurate or fabricated information in legal work and said its use required “caution and humility”.

Since then, AI companies and legal technology providers have continued to develop systems designed specifically for legal workflows.

Last week, OpenAI expanded its enterprise AI push into legal work with the launch of Astra for Law, a version of its GPT-6 Astra model configured for the industry.

The development came as reports of Anthropic considering a new model to respond to Astra’s early traction among enterprise users emerged. Notably, earlier this month, Thomson Reuters also announced that it is developing its own proprietary large language model (LLM) for legal and compliance work.

OpenAI’s Astra for Law will initially be available to selected law firms through its 'Trusted Access' programme and will later come to its application programming interface (API). Thomson Reuters, meanwhile, is already using its model within the workflow of CoCounsel Legal's Tabular Analysis, an AI-powered document review tool.

What is Astra for Law?

OpenAI introduced Astra for Law on September 17 as a new foundation for law firms and legal technology companies. The model combines GPT-6 Astra with legal-specific settings, tools, context and a legal search index.

In a blog, OpenAI said API customers including Harvey and Legora will be able to build products and workflows using the model. The company is also offering 26 ecosystem plugins that connect ChatGPT with specialist tools used by law firms, including Relativity and Clio.

The legal search system is designed to find US case law, statutes, regulations, court rules and administrative decisions. “The index covers more than 230 million URLs, with sources added daily. Its work with Free Law Project also gives the system access to a case-law collection covering more than 99.9 per cent of published US precedential case law,” OpenAI said.

How does Astra for Law perform on legal research?

OpenAI tested Astra for Law on 200 US legal research questions from the private validation set of Vals AI’s Legal Research Bench.

At the highest reasoning setting, OpenAI said 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 described this as a 40 per cent relative improvement.

On case-law questions, OpenAI said Astra for Law found 24 per cent more reference cases than GPT-6 Astra with web search alone. On an audited set of target passages, it retrieved up to 54 per cent more relevant passages from the correct court opinions.

“The system is also configured for legal analysis and writing. It can work with client facts, develop arguments or deal terms, and identify weaknesses and areas of uncertainty,” OpenAI said.

The company said Astra for Law can, for example, distinguish a court's holding from other observations and identify cases that weaken an argument.

What Anthropic has planned

Anthropic is considering launching a new AI model ahead of its IPO as it faces competitive pressure from OpenAI's GPT-6 Astra, according to a Reuters report. The company has not reportedly made a final decision on the release.

Anthropic is also evaluating the safety of its next model as part of discussions about a possible release, the report added.

The company is reportedly weighing investment in new models against profitability as investors focus more closely on when AI companies can generate sustainable profits.

Why Astra has increased pressure on Anthropic

OpenAI released GPT-6 Astra on September 3, with a focus on computer use, software engineering, cybersecurity and professional work. Reuters reported that the model has received a strong response from enterprise users.

According to Reuters, data from corporate expense platform Ramp showed Astra accounting for about 13 per cent of enterprise AI spending, compared with about 8 per cent for Anthropic's Claude Fable.

OpenAI models also led Anthropic's models in user spending on OpenRouter last week, according to the platform. Reuters reported that this was the first time OpenAI had led on that measure in more than two and a half years.

Anthropic still has a larger reported annualised revenue run rate. Reuters said it exceeded $65 billion by the end of July, compared with more than $40 billion for OpenAI.

Thomson Reuters is building its own legal AI model

OpenAI and Anthropic are not the only companies developing specialised AI for legal work. Thomson Reuters has built Thomson, its proprietary LLM for high-stakes professional work, including legal and compliance applications.

Thomson Reuters said the model uses its proprietary content, including Westlaw and Practical Law, alongside input from its network of legal experts. The company said owning more of the AI stack gives it greater control over performance, economics, deployment and data protection.

The company said its decision to build the model began in 2024 after it acquired Safe Sign Technologies, an AI research company founded by AI and legal experts. Thomson is built on open-source foundations that Thomson Reuters said it can continue to evolve.

Thomson Reuters said the model's first implementation powers CoCounsel Legal's Tabular Analysis feature. The tool can extract structured answers from up to 10,000 documents across up to 100 questions, with citation-grounded outputs.

The company said CoCounsel had crossed one million users as of the first quarter of 2026, while the share of its business that is generative-AI enabled had doubled in a little over a year.