OpenAI unveiled its newest language‑model family, GPT‑5.6, on Friday, a move that came less than 24 hours after the company announced it would stagger the rollout at the behest of the Trump administration. The suite comprises three distinct models: Sol, the flagship offering; Terra, a mid‑tier system designed for high‑volume tasks; and Luna, a fast, low‑cost option aimed at everyday users.
Pricing reflects the tiered approach. Sol charges $5 for every million input tokens and $30 for every million output tokens, roughly half the cost of Anthropic’s Claude Fable 5. Terra’s rates sit at half Sol’s price, while Luna is priced at less than half of Terra’s. OpenAI also introduced two specialized modes for Sol: a “max” setting that deepens reasoning capabilities and an “ultra” mode that leverages sub‑agents to handle complex, multi‑step problems.
Safety dominates the announcement. OpenAI says GPT‑5.6 is trained to refuse any request that involves prohibited cyber assistance, even when users attempt to disguise their intent or jailbreak the model. According to the company, Sol is better at identifying and fixing vulnerabilities than it is at carrying out end‑to‑end attacks, and it does not cross the “cyber‑critical” threshold defined in OpenAI’s updated preparedness framework.
To back those claims, OpenAI devoted roughly 700,000 A100e GPU hours to automated red‑team testing and partnered with external testers for an additional two‑week evaluation period. The company stresses that its “most robust safety stack to date” includes strengthened protections for high‑risk activities, sensitive cyber requests, and repeated misuse.
The preview phase will run under close government scrutiny. The Trump administration will review each user request on a case‑by‑case basis, a process OpenAI described as a short‑term measure to secure broader access later. The company warned that safeguards might occasionally block legitimate work, especially in dual‑use scenarios where defensive and offensive actions appear similar.
OpenAI expects the GPT‑5.6 suite to become generally available within weeks, citing a commitment to “broad access” while it continues to cooperate with U.S. officials on an emerging cyber executive order framework. The firm added that it does not view the current level of government involvement as a model for future releases, arguing that permanent oversight could limit the tools available to developers, enterprises, and security professionals.
Industry observers note that the rapid launch, coupled with explicit safety guarantees, marks a shift in how leading AI firms navigate regulatory pressure. By aligning the rollout with government expectations, OpenAI hopes to set a precedent for responsible deployment without sacrificing the commercial momentum that has defined its recent product cycles.
Este artículo fue escrito con la asistencia de IA.
News Factory APP - noticias agénticas para impulsar tu SEO y AEO.