OpenAI Lists Pricing for GPT-6 Astra for Law, a Legal Configuration of Its Flagship Model, Ahead of Any Official Announcement

OpenAI is readying a legal-focused version of its flagship model. Since Saturday, October 31, its official API pricing page has displayed a new section labeled “Legal model configurations,” featuring a single entry: gpt-6-astra-law, described in an accompanying note as “GPT-6 Astra for Law.” It is priced at $10 per million input tokens, $50 for output, and $1 for cached input, with “promotional pricing” guaranteed “at least through December 31, 2026.” The listing has appeared without any formal announcement.

What the Page Shows

The section is described as OpenAI’s “latest legal model configurations.” The wording matters: the page refers not to a brand-new model, but to a configuration of GPT-6 Astra. Furthermore, the listed prices are identical to standard processing rates for gpt-6-astra on the very same page: $10, $50, and $1 per million tokens. In other words, the legal version’s “promotional” price currently equals the base model’s cost, carrying no premium.

As of this writing, around 11:00 AM Paris time, there is no blog post, no model card, and no API changelog entry: the URL where OpenAI typically hosts model documentation returns a 404 error. The entry was first spotted around 9:00 AM Paris time, according to AI Tracker Bot on X, an account that tracks changes to AI labs’ pages.

A Vertical Product Line Takes Shape

This new entry is not an isolated experiment. The pricing page already reserves dedicated blocks for two domain-specific families: Daybreak cybersecurity models (including gpt-5.6-cyber) and Rosalind life sciences models (including gpt-6-rosalind), with the latter restricted to approved research via OpenAI’s trusted access program. Law has now become the third specialized domain to receive its own pricing tier.

The legal sector is an established market where vertical software developers have built solutions on top of foundation models for several years. A “for Law” package sold directly by OpenAI at parity with base model pricing could either challenge these vendors directly or serve as an underlying layer for their tools.

What Remains Unknown

Beyond pricing, virtually everything else is up in the air. OpenAI has not explained what a “configuration” entails—whether it involves tailored system instructions, dedicated legal research tools, strict citation parameters, or actual retraining. Access requirements also remain unspecified: it is unclear whether the model will be open to all developers or gated behind a vetting program like Rosalind, which jurisdictions or legal systems it targets, and how it performs, as no benchmark results have been shared. Finally, an unannounced pricing line provides no guarantee of a release date or whether the final product name will remain unchanged.