Estimated reading time: 5 minutes · Last updated:
OpenAI launched Astra for Law on Sept. 17, 2026, a legal-focused foundation that pairs its GPT-6 Astra model with settings, tools and context tailored for law firms and legal-technology vendors. The company says Astra for Law includes 26 new plugins that connect ChatGPT to specialized legal tools and will be available initially to selected law firms through Trusted Access in ChatGPT and Codex, with an API rollout planned soon. OpenAI reported that, versus GPT-6 Astra using web search alone, Astra for Law was 40% more accurate, produced 24% more reference cases and retrieved 54% more relevant passages from correct court opinions.
Key takeaways
- Product launch: OpenAI launched Astra for Law on Sept. 17, 2026, built on GPT-6 Astra with legal settings and tools.
- Plugins: OpenAI added 26 new ecosystem plugins that connect ChatGPT to specialized legal tools.
- Performance claims: OpenAI says Astra for Law was 40% more accurate, found 24% more reference cases and retrieved 54% more relevant passages versus GPT-6 Astra using web search alone.
- Market context: PYMNTS reported that funding to legal technology startups surpassed $2.4 billion in the first nine months of 2025.
Table of contents
What Astra for Law is and how it is built
Astra for Law is an OpenAI foundation tailored to legal workflows that combines the company’s latest frontier model, GPT-6 Astra, with curated settings, tools and contextual prompts aimed at professional legal work. The package is not just a model release: OpenAI pairs GPT-6 Astra with legal context and tool access so firms can configure outputs for research, contracting and other practice-area tasks.
The product adds 26 ecosystem plugins that link ChatGPT to specialized legal software, allowing customization around a firm’s existing toolset. OpenAI says it will extend these legal capabilities to future frontier models as they ship, keeping the legal layer aligned with model upgrades.
How OpenAI measured Astra for Law’s performance
OpenAI published comparative figures that test Astra for Law against GPT-6 Astra using web search alone. The company reported that Astra for Law was 40% more accurate in providing research answers, that it produced more comprehensive answers, that it found 24% more reference cases and that it retrieved 54% more relevant passages from the correct court opinions. Those three percentage figures are the core performance claims OpenAI presented for legal research tasks.
The figures come from OpenAI’s announcement and are framed as results of internal evaluations; OpenAI also said it will continue to refine models, settings and tools through rigorous evaluations and feedback from lawyers and legal-technology partners. Law firms and vendors that consider adoption will need to test the claims against their own benchmarks and document types before replacing established research workflows.
Distribution, integration and early availability
OpenAI will initially offer Astra for Law to selected law firms via Trusted Access in ChatGPT and Codex, with a promise to add the product to its API soon. The Trusted Access route signals a phased rollout: firms in the pilots can test the product inside ChatGPT and Codex before broader API-based integrations are available.
The 26 plugins connect ChatGPT to specialized tools used by law firms, which is intended to let firms bring their document stores, practice-management systems and other software into Astra for Law workflows. OpenAI framed this as enabling firms to build around their expertise and connect the tools they use, while it iterates on instructions and settings with partner feedback.
Market context and competitors in legal AI
OpenAI’s launch arrives as AI adoption in law moves from pilots toward production. PYMNTS reported in October 2025 that AI was performing research, contracting, compliance and billing, and that funding to legal-technology startups surpassed $2.4 billion in the first nine months of 2025. Those trends help explain why OpenAI is investing directly in a legal layer for its frontier models.
Anthropic is an explicit competitor cited in the broader coverage: in May 2026 Anthropic added several tools to Claude for the legal industry, including more than 20 Model Context Protocol connectors and 12 plugins, and said that legal professionals had become the most engaged Claude Cowork users of any knowledge-work function. That parallel investment in connectors and plugins frames the legal market as one where integration with firm tooling matters as much as raw model capability.
| Product | Model | Plugins / Connectors | Initial availability |
|---|---|---|---|
| Astra for Law | GPT-6 Astra | 26 plugins | Selected law firms via Trusted Access in ChatGPT and Codex |
| Anthropic (Claude updates) | Claude | More than 20 MCP connectors and 12 plugins | Updates announced in May 2026 |
How Astra for Law could gain or lose traction
The case for
- Firms that value tighter integrations will be able to link Astra for Law to existing systems through the 26 plugins, reducing friction for pilots and proofs of concept.
- OpenAI’s commitment to iterate models, settings and instructions with lawyer feedback increases the chance that accuracy and relevance will improve over time.
The case against
- OpenAI’s performance claims are based on its own evaluations; firms must validate the 40%, 24% and 54% figures on their document mixes and jurisdictions before relying on the tool for critical research.
- Competitors such as Anthropic are already offering connectors and plugins, making integration breadth and vendor relationships a deciding factor beyond raw model accuracy.
What to be careful about
- OpenAI’s accuracy figures come from its announcement and require independent validation on firm-specific corpora and jurisdictional case law.
- Integration risk: plugins must map reliably to firm systems and permissions models, or firms risk operational gaps when moving from pilot to production.
- Competitive risk from vendors that already provide connectors and practice-focused tooling could slow enterprise adoption of Astra for Law.
The bottom line
OpenAI’s Astra for Law packages GPT-6 Astra with legal settings and 26 plugins to make the model more usable inside law-firm workflows, and the company has framed the release as the start of a longer-term investment. The announcement includes concrete performance claims — 40% more accurate, 24% more reference cases and 54% more relevant passages — but those numbers are OpenAI’s internal results and will need independent validation on firm-specific datasets. Adoption will hinge on how reliably the plugins integrate with firm systems and on how quickly OpenAI and competitors expand connectors across practice-area software.
What to watch
- Watch for Astra for Law to appear on OpenAI’s API; no date has been set.
- Watch for pilots to expand beyond selected law firms as OpenAI collects feedback; no date has been set.
- Watch for OpenAI to roll legal capabilities into subsequent frontier models; no date has been set.
Frequently asked questions
What exactly is Astra for Law?
Astra for Law is an OpenAI foundation that pairs GPT-6 Astra with legal-specific settings, tools and 26 plugins designed to connect ChatGPT to specialized legal software and support practice workflows.
How did OpenAI measure Astra for Law’s results?
OpenAI reported that Astra for Law was 40% more accurate than GPT-6 Astra using web search alone, produced 24% more reference cases and retrieved 54% more relevant passages from correct court opinions; those figures come from OpenAI’s announcement.
How can law firms access Astra for Law now?
OpenAI said it will initially offer Astra for Law to selected law firms through Trusted Access in ChatGPT and Codex, and that it plans to add the product to its API soon.
Related reading
