OpenAI is rolling out GPT-5.6, but only to some users because of the Trump administration

OpenAI on Friday unveiled its GPT-5.6 family of AI models, launching Limited preview Its new models are codenamed Sol, Terra and Luna ahead of the wider planned release expected in the coming weeks.

The announcement comes one day after reports that the administration of President Donald Trump asked OpenAI to limit the initial release of GPT-5.6 while officials evaluate the model within an evolving federal framework for frontier AI systems, reports that have proven accurate.

Calling it a flagship model, OpenAI said GPT-5.6 Sol improves performance across programming, biology and cybersecurity. Terra offers performance similar to GPT-5.5 at a lower cost, while Luna is designed for high-volume, low-cost workloads, the company said. The release also introduces “Extreme” and “Ultra” thinking modes, giving Soul more time to solve complex problems or coordinate multiple sub-agents for difficult tasks.

“We have begun a limited preview of the GPT‑5.6 series: Sol, our flagship model; Terra, a balanced model for everyday work; and Luna, a fast and affordable model,” OpenAI wrote. “Terra has competitive performance GPT-5.5 While it is twice as cheap, Luna offers powerful capabilities at our lowest cost.

In the announcement, OpenAI confirmed that it had shared models with the US government ahead of the launch, and had begun a limited preview at the administration’s request while the two sides develop a process for future frontier AI releases.

“As part of our ongoing engagement with the US government, we previewed our plans and model capabilities ahead of today’s launch,” the company wrote. “At their request, we began with a limited preview to a small group of trusted partners whose engagement was shared with the government, ahead of a wider rollout.”

The move follows the administration’s previous order that Anthropy Limit access she has Myth 5 and Myths 5 This makes GPT-5.6 the second frontier AI system this month whose rollout has been affected by the White House.

In testing, OpenAI claimed that Sol achieved the highest scores in TerminalBench, a benchmark for command-line software engineering tasks, and outperformed GPT-5.5, Cloud Mythos 5, and Fable 5.

“GPT-5.6 Sol also shows broad improvements in biology workflows,” OpenAI wrote. “In the first version of GeneBench, which evaluates long-term genomics and quantitative biological analyses, it achieves stronger results than GPT-5.5 while using fewer codes.”

On cybersecurity, OpenAI said GPT-5.6 combines stronger cybersecurity capabilities with expanded safeguards designed to support defensive security research, while limiting offensive misuse. OpenAI added that the model still falls below its cyber criticality threshold because although it can identify vulnerabilities and exploit components, it cannot independently produce a complete exploit chain during testing.

“GPT-5.6 is trained to reject blocked cyber assistance, including when users attempt to hide their intentions or jailbreak a form,” the company said. “These model-level safeguards set the first boundaries around what the model should and should not help with.”

According to OpenAI, during the preview, GPT-5.6 Sol, Terra, and Luna will be available through the API and Codex Alimentarius To a select group of partners before expanding generally to ChatGPT and other users. OpenAI also introduced a new naming system for the model family and said GPT-5.6 Sol will launch on Cerebras in July, offering inference speeds of up to 750 characters per second.

Despite the limited rollout required, OpenAI said it still intends to make the models widely available to the public.

“We believe in broad accessibility, and plan to make GPT‑5.6 Sol, Terra, and Luna generally available in the coming weeks,” they said.