In short
- OpenAI has released partial previews of GPT-5.6 Sol, Terra, and Luna ahead of the larger release.
- The company said the restricted release follows a request from the US government as officials develop a system to monitor advanced AI models.
- GPT-5.6 adds new concepts, stronger capabilities for online security, and increased protection against misuse.
OpenAI on Friday unveiled the GPT-5.6 family of AI models, introducing a limited display of its new Sol, Terra, and Luna models ahead of a wider release expected in the coming weeks.
The announcement comes a day after reports about President Donald Trump’s administration Asked OpenAI to reduce the initial release of GPT-5.6 while officials are evaluating the model under the government’s policy on the limits of AI systems-reports confirmed.
Calling it a popular version, OpenAI said GPT-5.6 Sol will improve performance in coding, biology, and cybersecurity. The company said that Terra offers the same functionality as GPT-5.5 at a lower price, while Luna is designed for a higher, more affordable price. This release also introduces “max” and “ultra” thinking modes, which give Sol more time to solve complex problems or coordinate multiple subagents for complex tasks.
“We are starting a small preview of the GPT-5.6 series: Sol, our top model; Terra, a suitable model for daily tasks; and Luna, a fast and affordable model,” OpenAI wrote. “Terra has enough competition GPT-5.5 being 2x cheaper, and Luna brings power at a much lower price. “
In the announcement, OpenAI confirmed that it had shared these samples with the US government before it was launched, and it begins with a preview of what the administration requested as the two sides develop a way to push the boundaries of AI for the future.
The company wrote: “As part of our partnership with the US government, we previewed our plans and capabilities before today’s launch. “At their request, we are starting with a small preview of a small group of trusted people whose participation has been shared by the government, before releasing more information.”
This move follows the order he previously ordered Anthropic reduce the chances to his own Myth 5 is Mythos 5 examples, making GPT-5.6 the second AI system this month whose release has been touched by the White House.
In testing, OpenAI said Sol achieved the highest score on TerminalBench, a benchmark for software engineering, outperforming GPT-5.5, Claude Mythos 5, and Fable 5.
“GPT-5.6 Sol also marks a significant change in the flow of biology,” OpenAI wrote. “On GeneBench v1, which evaluates extended genomics and quantitative-biology analysis, it achieves stronger results than GPT-5.5 using fewer tokens.”
In terms of cybersecurity, OpenAI said GPT-5.6 pairs strong cyber security capabilities with additional safeguards designed to support security research, and reduce misuse. OpenAI added that the model remains under cyber Critical because, although it can detect vulnerabilities and exploit certain features, it cannot produce a complete technology in testing.
“GPT-5.6 is trained to resist illegal cyber support, including when users want to hide their data or violate a model,” the company said. “The brand’s defenses set the first boundaries around what the brand should and shouldn’t support.”
According to OpenAI, in preview, GPT-5.6 Sol, Terra, and Luna will be available via API and Codex to a selected group of friends before adding publicly to ChatGPT with other users. OpenAI also introduced a new model family naming scheme and said GPT-5.6 Sol will be launched on Cerebras in July, offering speeds of up to 750 tokens per second.
Despite being ordered to be released, OpenAI said it still wants the models to be available to the public.
“We believe in accessibility, and we want to make GPT-5.6 Sol, Terra, and Luna available in the coming weeks,” he said.
Daily Debrief A letter
Start each day with top stories right here, including originals, podcasts, videos and more.
