OpenAI has unveiled its next-generation artificial intelligence model, GPT-5.6, but has decided to limit its initial release to select institutions at the request of the U.S. government.

OpenAI announced on June 26 (local time) that it has developed the new GPT-5.6 model in three distinct versions: Sol, Terra, and Luna.

Among them, the top-tier Sol model is described by OpenAI as possessing the highest performance to date and autonomous agent capabilities in fields such as coding, biology, and cybersecurity.

To enhance the model's reasoning capabilities, OpenAI introduced a "maximum reasoning effort" option, which allows the model to dedicate the longest processing time ever, and an "Ultra mode" that utilizes sub-agents to perform complex tasks more quickly.

OpenAI also emphasized that on the "Terminal-Bench 2.1" performance benchmark, which measures terminal coding ability, GPT-5.6 Sol Ultra scored 91.9% and GPT-5.6 Sol scored 88.8%, surpassing the 88% achieved by Anthropic's Claude Mythos 5.

In particular, OpenAI highlighted that the Sol model is equipped with the most robust safety system to date, strengthening protections against sensitive cyber requests and misuse that could lead to high-risk activities or hacking.

OpenAI did not disclose the detailed performance metrics that it has typically provided with previous model announcements.

The mid-tier model, Terra, offers performance similar to its predecessor, GPT-5.5, while cutting the price in half, and the entry-level model, Luna, provides decent performance at the lowest cost, the company said.

OpenAI stated that it would first provide these models to the government and "trusted partners," noting that a general public release would take place in a few weeks.

However, OpenAI pointed out, "We do not believe this type of government approval process should become a long-term standard," adding that it "prevents users, developers, companies, cybersecurity experts, and global partners who need these tools from utilizing the best technology."

Earlier this month, President Trump signed an executive order requiring AI companies to submit new models to the government up to 30 days before their release. Additionally, the U.S. Department of Commerce issued export control guidelines, stating that Anthropic's advanced models, Mythos 5 and Fable 5, pose national security risks.

Meanwhile, it has been reported that OpenAI is delaying its initial public offering (IPO), which was originally expected to take place this year.

OpenAI, which submitted a confidential IPO filing to the Securities and Exchange Commission (SEC) earlier this month, has not yet held preliminary meetings to discuss potential offering prices or demand, nor has it presented an official schedule for the listing, CNBC reported on this day, citing sources.

Previously, The New York Times (NYT) reported on June 26 that OpenAI is leaning toward postponing its IPO, which it had aimed to pursue within the year, to next year.

(Photo: AP, Yonhap News)