The U.S. government has cleared OpenAI to broadly release its new GPT-5.6 model family, ending a restricted preview that limited access to vetted partners. The decision marks a significant milestone in Washington’s emerging oversight framework for frontier AI systems. Initially delayed last month due to security concerns, the rollout will now proceed with all three variants—Sol (flagship), Terra (mid‑tier), and Luna (fastest/cost‑efficient)—expected to become publicly available on Thursday.
The approval came after additional testing and consultations between OpenAI and officials. The Department of Commerce, which had previously imposed the pause, confirmed that the models met revised safety benchmarks. GPT-5.6 Sol is described as OpenAI’s most capable system yet, with improvements in coding, scientific reasoning, biology workflows, and cybersecurity tasks. It also introduces a higher‑reasoning setting and an advanced mode for coordinating sub‑agents on multi‑step tasks.
OpenAI initially restricted access to approved organizations through the API and Codex, excluding ChatGPT from the preview. The company emphasized that the government‑driven process should not become the long‑term norm but cooperated to develop a repeatable blueprint for future releases. The green light follows the reinstatement of Anthropic’s Fable and Mythos models, which had faced similar three‑week holds over the potential misuse of advanced vulnerability‑discovery capabilities.
The broader deployment is set to intensify competition in enterprise AI, software engineering, and cybersecurity markets, while also raising barriers for smaller labs that lack resources for structured safety reviews. Terra will be priced at half the cost of GPT-5.5, a move aimed at capturing a wider customer base. As Washington continues to define technical thresholds for frontier‑model security restrictions, the GPT-5.6 approval sets a precedent that may shape how future advanced AI systems enter the commercial market.
