OpenAI unveiled its next-generation artificial intelligence (AI) model, "GPT-5.6." However, at the request of the U.S. government, it will be provided first to select institutions and partners for the time being.
OpenAI said on the 26th (local time) that it developed GPT-5.6 in three models—Sol, Terra and Luna. Among them, the top-tier model "Sol" improves performance in specialized areas such as coding, biology and cybersecurity, and strengthens agent capabilities that autonomously carry out complex tasks.
The new model also introduces a maximum inference effort option that grants longer reasoning time and an Ultra mode that uses multiple sub-agents to speed up tasks. OpenAI said GPT-5.6 Sol Ultra scored 91.9% and GPT-5.6 Sol scored 88.8% on "Terminal-Bench 2.1," which evaluates terminal coding ability.
In this announcement, OpenAI did not widely disclose granular performance metrics as it did with previous model releases, but said it strengthened safeguards for the Sol model, including blocking cyber requests that involve high-risk activities or could be exploited for hacking.
The mid-tier model, Terra, was introduced as maintaining performance similar to GPT-5.5 while halving the price. The lower-tier model, Luna, focused on providing basic performance at low expense.
OpenAI plans to first provide GPT-5.6 to the government and trusted partners under shared agreements, then roll out to general users and developers a few weeks later.
However, the company also said that a restricted distribution approach based on government approval should not become a long-term standard. OpenAI noted, "Necessary users and developers, corporations, cybersecurity experts, and global partners may be unable to use the latest tools."