OpenAI lists Astra for Law API pricing, with access still limited
OpenAI’s API pricing page now lists gpt-6-astra-law at $10 for input, $50 for output and $1 for cached input per million tokens. The listing was visible on October 31, but the page does not establish when the row was added.
The important distinction is access. OpenAI’s current help guide still describes the API as forthcoming. A published rate card is not evidence that every developer account can call the configuration today.
The published token rates
| Token category | Price per million tokens |
|---|---|
| Input | $10 |
| Output | $50 |
| Cached input | $1 |
OpenAI says these promotional rates remain available at least through December 31, 2026. That wording does not specify a firm expiry date or a later price. The row appears under legal model configurations, rather than announcing a separate foundation model.
Who can get access
The help guide identifies selected US law firms as the initial audience, through Trusted Access and Codex. Eligibility covers lawyers and people working under their supervision. Firms seeking early access are directed to their OpenAI account team or sales.
That is narrower than an ordinary self-service API release. Before planning an integration, a team needs confirmation that its organization is eligible and that the required access surface is enabled. The price listing answers a cost question, not those deployment questions.
What the legal configuration adds
Astra for Law was introduced on September 17. OpenAI describes a combination of GPT-6 Astra, a dedicated legal search index and instructions for legal analysis and writing. It is intended to connect model reasoning with relevant legal material rather than rely on a generic prompt alone.
The index covers US case law, statutes, regulations, court rules and administrative decisions, with new sources added daily. OpenAI’s guide tells users to review answers and cited material before relying on them. Neither the product name nor the price establishes that a particular answer is correct.
OpenAI’s announcement includes litigation and transactional examples. These illustrate the intended workflow: locate relevant authority, relate it to the supplied facts and surface weaknesses. They are vendor-selected examples, not a ByteForward test or a guarantee of results on another matter.
What to verify before building
Keep three checks separate: permission to use the configuration, the product surface your team can access and the review process for its output. An API budget does not replace any of them. For a firm evaluating the offering, the useful next step is to reconcile the pricing page with its own access confirmation, rather than treat a new row as a launch notice.
For separate research context, see OpenAI’s Ironclad contracting evaluation. That study tested software tasks, not general access to Astra for Law.
