In short
- OpenAI launched a restricted preview of GPT-5.6 Sol, Terra, and Luna forward of a broader deliberate launch.
- The corporate mentioned the restricted rollout follows a request from the U.S. authorities whereas officers develop a framework for evaluating superior AI fashions.
- GPT-5.6 provides new reasoning modes, stronger cybersecurity capabilities, and expanded safeguards towards misuse.
OpenAI on Friday unveiled the GPT-5.6 household of AI fashions, launching a restricted preview of its new fashions codenamed Sol, Terra, and Luna earlier than a broader deliberate launch anticipated within the coming weeks.
The announcement comes someday after experiences that President Donald Trump’s administration requested OpenAI to restrict GPT-5.6’s preliminary launch whereas officers consider the mannequin underneath a creating federal framework for frontier AI techniques—experiences that proved correct.
Calling it the flagship mannequin, OpenAI mentioned GPT-5.6 Sol improves efficiency throughout coding, biology, and cybersecurity. The corporate mentioned Terra delivers efficiency similar to GPT-5.5 at a decrease value, whereas Luna is designed for high-volume, low-cost workloads. The discharge additionally introduces “max” and “extremely” reasoning modes, giving Sol extra time to unravel complicated issues or coordinate a number of subagents for demanding duties.
“We’re starting a restricted preview of the GPT‑5.6 collection: Sol, our flagship mannequin; Terra, a balanced mannequin for on a regular basis work; and Luna, a quick and reasonably priced mannequin,” OpenAI wrote. “Terra has aggressive efficiency to GPT‑5.5 whereas being 2x cheaper, and Luna brings robust functionality at our lowest value.”
Within the announcement, OpenAI confirmed it shared the fashions with the U.S. authorities earlier than launch, and is starting with a restricted preview on the administration’s request whereas the 2 sides develop a course of for future frontier AI releases.
“As a part of our ongoing engagement with the U.S. authorities, we previewed our plans and the fashions’ capabilities forward of immediately’s launch,” the corporate wrote. “At their request, we’re beginning with a restricted preview for a small group of trusted companions whose participation has been shared with the federal government, earlier than releasing extra broadly.”
The transfer follows the administration’s earlier order that Anthropic restrict entry to its Fable 5 and Mythos 5 fashions, making GPT-5.6 the second frontier AI system this month whose rollout has been impacted by the White Home.
In testing, OpenAI claimed Sol achieved the best rating on TerminalBench, a benchmark for command-line software program engineering duties, and outperformed GPT-5.5, Claude Mythos 5, and Fable 5.
“GPT‑5.6 Sol additionally exhibits broad enhancements in biology workflows,” OpenAI wrote. “On GeneBench v1, which evaluates long-horizon genomics and quantitative-biology analyses, it achieves stronger outcomes than GPT‑5.5 whereas utilizing fewer tokens.”
When it comes to cybersecurity, OpenAI mentioned GPT-5.6 pairs stronger cybersecurity capabilities with expanded safeguards designed to assist defensive safety analysis, whereas limiting offensive misuse. OpenAI added that the mannequin stays beneath its Cyber Essential threshold as a result of, though it could establish vulnerabilities and exploit parts, it couldn’t autonomously produce a whole exploit chain throughout testing.
“GPT‑5.6 is skilled to refuse prohibited cyber help, together with when customers try to disguise their intent or jailbreak the mannequin,” the corporate mentioned. “These model-level safeguards set up the primary boundary round what the mannequin ought to and mustn’t assist with.”
Based on OpenAI, throughout the preview, GPT-5.6 Sol, Terra, and Luna will probably be accessible by way of the API and Codex to a choose group of companions earlier than increasing publicly to ChatGPT and different customers. OpenAI additionally launched a brand new naming system for the mannequin household and mentioned GPT-5.6 Sol will launch on Cerebras in July, providing inference speeds of as much as 750 tokens per second.
Regardless of the ordered restricted rollout, OpenAI mentioned it nonetheless intends to make the fashions broadly accessible to the general public.
“We consider in broad entry, and we plan to make GPT‑5.6 Sol, Terra, and Luna typically accessible within the coming weeks,” they mentioned.
Each day Debrief E-newsletter
Begin daily with the highest information tales proper now, plus unique options, a podcast, movies and extra.