OpenAI launches new models but only for ‘trusted partners’

OpenAI announced the launch of GPT -5.6, its new family of artificial intelligence models. At the request of the Trump administration, access is initially limited to a select group of White House-approved partners. The move follows a dynamic already seen with GPT-5.5. Three models: Solthe flagship model for complex tasks and advanced reasoning; Terrathe mid-range version; Lunathe fastest and cheapest. All three are classified as “high risk” in the cybersecurity categories, but none meet the critical threshold required by the company’s internal security framework. They are capable of identifying vulnerabilities in computer systems, but not of conducting end-to-end autonomous attacks against well-organized targets.

The tests also detected new behavior in the Sol and Terra models: compared to GPT-5.5, they tend to perform actions that the user has not explicitly requested to a greater extent. In the new naming scheme, the number identifies the generation, while Sol, Terra and Luna indicate distinct levels of ability, each with its own developmental trajectory. Prices vary per million tokens: Sol at 5 dollars in input and 30 in output, Earth at 2.50 and 15, Luna at 1 and 6.

OpenAI follows Trump’s directives

The launch comes at a time of strong regulatory tension. In recent weeks, the Trump administration has banned foreign entities from accessing two Anthropic models, Fable 5 and Mythos 5, forcing the company to completely suspend access to comply with the rule. The stalemate has now lasted for two weeks.

According to the Wall Street Journal, critics of the sector accuse the White House of favoring OpenAI – whose leaders have politically supported Trump – by penalizing the Amodei brothers’ companywith whom relations have been more conflictual. The Commerce Department, through the Center for AI Standards and Innovation, is reviewing GPT-5.6 as it has done for other models. A recent executive order from Trump expanded the role of national security agencies in this assessment process.

OpenAI: “It must not become the norm”

OpenAI said it hoped for general availability in the coming weeks, noting that the government prior approval mechanism is not expected to become the norm. “We are keeping the best tools away from the users, developers, companies, cybersecurity experts and international partners who need them,” the company wrote, calling the move a transitional phase as it waits for Trump’s executive order on oversight of AI models to go into effect.