OpenAI Begins US-Only Preview of GPT-5.6 After Government Request

OpenAI on Friday announced the limited release of its GPT-5.6 series of artificial intelligence models, available initially only to a select group of US-based partners. The company stated that the preview was conducted at the request of the US government, following a briefing on the models' capabilities.

The launch comes two weeks after the US government directed OpenAI's rival Anthropic to restrict access to its Fable 5 and Mythos 5 models exclusively to US nationals, citing national security concerns. That move surprised many in the technology industry, given the current administration's broader push to reduce AI oversight and block states from creating their own regulations.

The GPT-5.6 series includes three models: Sol, positioned as OpenAI's new flagship; Terra, a mid-range model intended for everyday tasks; and Luna, a fast and cost-effective option. OpenAI said Terra would be priced at half the cost of its predecessor, GPT-5.5, as the company seeks to retain customers amid rising competition from Anthropic and Google.

According to OpenAI, the company shared details of its new models with US authorities before the launch and, at the government's request, is initially offering access only to a set of trusted partners whose identities have been disclosed to officials. Although the partners are US-based, OpenAI noted that overseas employees of those companies or entities will also be able to use the new models.

Both Anthropic's Mythos models and OpenAI's GPT-5.6 series have raised concerns due to their reportedly advanced ability to identify software vulnerabilities—weaknesses in code that hackers can exploit. In response to these concerns, President Donald Trump earlier this month signed an executive order establishing a voluntary federal review process for national security risks in advanced AI models before their public release.

The developments highlight the growing tension between the rapid advancement of AI capabilities and government efforts to manage potential security risks. While the White House has generally favoured a hands-off approach to AI regulation, the novel features of these models have prompted targeted interventions. The limited preview of GPT-5.6 reflects a collaborative approach between industry and government, though it remains to be seen whether broader restrictions may follow.