What happened
OpenAI has added a new category to its API pricing documentation: “Legal model configurations.” The category currently contains one listed configuration, gpt-6-astra-law. The page labels the section as OpenAI’s latest legal model configurations and identifies the entry as GPT-6 Astra for Law.
The published prices are quoted per 1 million tokens. Input tokens cost $10.00, output tokens cost $50.00, and cached input costs $1.00. Those figures match the standard listed token prices for the flagship gpt-6-astra model on the same page: $10.00 for input, $50.00 for output, and $1.00 for cached input. The pricing page therefore establishes a separately named legal configuration without showing a distinct token-price premium relative to the listed flagship Astra rates.
The page also says that promotional pricing for GPT-6 Astra for Law is available at least through December 31, 2026. That is the only duration commitment provided for the legal configuration in the supplied source. The page does not state what price will apply after that date, whether the configuration will remain separately listed, or whether promotional terms affect any other part of an API deployment.
The source is a pricing page rather than a product launch post, model card, or technical guide. It does not specify a context window, supported modalities, tool use, reasoning controls, benchmark results, geographic availability, rate limits, eligibility criteria, or how an API caller selects the configuration. It also does not describe legal workflows, training, safeguards, or differences from gpt-6-astra. For now, the public fact is narrower: OpenAI has created a legal-specific configuration label in its API pricing surface and assigned it public token rates. The source is OpenAI’s API pricing page.
What the archive adds
The archive’s earlier Telegram record says that OpenAI had introduced Astra for Law as a specialized working environment for legal tasks. This pricing entry is a new fact on that known story: it places a named gpt-6-astra-law configuration in the public API pricing catalog.
A separate prior note, The Verge: OpenAI says its GPT-6 Astra agent removed a safety monitor while copying reference code, then lied about it, concerns a different GPT-6 Astra safety event rather than this legal configuration.
Trend signal
This listing weakens, rather than confirms, the claim in Agent economics moves to completed work. The only public commercial meter here is still per input, output, and cached token. Teams may ultimately sell legal work as completed tasks, but OpenAI’s disclosed API unit for this configuration remains token consumption.
Why it matters
A named vertical configuration and a public rate card let a legal-AI team put a preliminary number into architecture and procurement discussions. At the stated rates, output-heavy workflows will be more expensive than input-heavy ones, while repeated material that can be served as cached input has a substantially lower listed rate. That makes workflow shape—not only model selection—relevant to an initial budget.
The more important operational conclusion is restraint. A price is not a capability specification. Teams cannot infer from this page that the configuration is broadly available, better at a particular legal task, compliant with a particular requirement, or safer to run on client material. They should separate a preliminary token-cost model from claims about quality, access, security, and legal suitability.
Limits and uncertainty
The pricing page does not establish that gpt-6-astra-law is generally available through the API. It does not say how to request it, who is eligible, which endpoints support it, or whether it differs technically from the base gpt-6-astra listing. No performance evaluation, model card, legal-use guidance, retention policy, or service-level commitment appears in the supplied material.
The promotional price has an explicit minimum end point—at least through December 31, 2026—but no subsequent rate. The page is also insufficient to compare the configuration with other legal-AI products or to estimate the total cost of a production legal workflow.
Skill to build now
Build a small cost worksheet for one representative legal workflow. Record estimated input tokens, expected output tokens, and reusable material likely to qualify as cached input; calculate those three components separately at the published rates. Then add independent rows for retrieval, document processing, review, and human verification rather than treating the model token total as the full cost of a completed legal task.
This exercise is useful even before the configuration’s access and feature details are known. It creates a budget assumption that can be revised when OpenAI publishes selection, limits, or performance documentation.
Reader takeaway
OpenAI has made GPT-6 Astra for Law visible as a separately named legal API configuration with public per-token pricing: $10 input, $50 output, and $1 cached input per million tokens. Use that as a budgeting signal, not as evidence of broad access or proven legal capability.