OpenAI to release GPT-5.6 after US security review

OpenAI was set to release GPT-5.6 to the public on Thursday after chief executive Sam Altman said the company made “many changes” during a security review involving US officials.

Altman told CNBC that US officials tested the model and raised concerns during the process. He described the review as “a collaborative back and forth.”

The company’s GPT-5.6 series has three tiers: Sol, its flagship model; Terra, a mid-range version for everyday work; and Luna, a faster, lower-cost option. Altman said Sol is “the best model in the world for most people” and is 54 percent more token efficient on tasks than competing models.

OpenAI said in late June that it had given preview access to GPT-5.6 to a limited group of trusted US-based partners at Washington’s request. The Trump administration gave the company clearance for a broad launch earlier this week after technical testing and meetings, while the White House said the voluntary scrutiny does not amount to formal approval.

The release comes as advanced AI models from OpenAI and rivals draw security concerns over their ability to find weaknesses in software code that hackers could exploit. Anthropic recently began restoring access to its most powerful models after Washington lifted a restriction on releasing them to foreigners, while Meta has also released new AI models aimed in part at coding and image generation.