Photo credit: ReutersOpenAI has unveiled GPT-5.6, its latest family of AI models, but unlike previous launches, the rollout is beginning under restrictions tied to a new US government review process. The company is introducing three models - GPT-5.6 Sol, GPT-5.6 Terra and GPT-5.6 Luna - with access initially limited to around 20 companies approved by the US government. While OpenAI says it expects broader availability in the coming weeks, it also acknowledged that the current approval process is a temporary measure as Washington develops a framework for releasing advanced AI systems.Key TakeawaysOpenAI has introduced GPT-5.6 Sol, Terra and Luna, with Sol positioned as the flagship model.Initial access is limited to around 20 government-approved companies.The company plans to expand availability next week, followed by a broader rollout in the coming weeks.GPT-5.6 adds new reasoning options, including an Ultra mode that can split work across multiple sub-agents.OpenAI says it is cooperating with the US government but does not view the current approval process as a long-term solution.GPT-5.6 arrives with three model variantsOpenAI's latest release consists of three separate models designed for different workloads. GPT-5.6 Sol is the most capable version in the lineup, while GPT-5.6 Terra is positioned as a balance between performance and efficiency. GPT-5.6 Luna, meanwhile, is aimed at users who prioritise speed and affordability.Alongside the new models, OpenAI is introducing additional reasoning capabilities. The company says GPT-5.6 includes options that allow for deeper reasoning as well as an "Ultra" mode that distributes work across multiple sub-agents. According to OpenAI, these additions are intended to improve how the models handle more complex requests while offering users greater flexibility depending on the task.Limited rollout follows government reviewUnlike previous OpenAI model launches, GPT-5.6 is not immediately available to all users. Instead, the company is beginning with a limited preview involving around 20 companies whose participation has been approved by the US government.OpenAI said it expects to expand access to additional organisations next week before moving toward a wider public rollout in the coming weeks. According to the company, the US government has been informed of its broader launch plans and has expressed support for expanding availability, provided no significant concerns emerge during the additional testing period.The company also said it has been giving government officials a sneak peek of GPT-5.6 over the last month, including conversations with CEO Sam Altman and the White House earlier in June. "OpenAI said it had anticipated a phased rollout, but not that initial access would require government approval for individual customers or be limited to roughly 20 partners.OpenAI says current process is temporaryOpenAI is working with the federal government during the rollout, but the company said it does not believe the current approval system should be the long-term model for releasing advanced AI systems.In its announcement, OpenAI said requirements for government access keep advanced AI tools out of the hands of developers, enterprises, cyber defenders, businesses and global partners who could benefit from them. But it also said it was ready to follow the temporary process as it believes it provides the strongest route to wider availability while policymakers finalise a repeatable framework for future releases.“It’s a time of transition,” OpenAI said.The US administration has announced plans to evaluate frontier AI models before wider deployment, but is still developing the detailed process that will govern those reviews. The company indicated it is working with the administration as that framework takes shape.Cybersecurity remains a major focusOne of the primary reasons behind the additional scrutiny is the growing cybersecurity capability of frontier AI models.OpenAI said its internal assessment found that GPT-5.6 Sol is better suited to helping users identify and fix software vulnerabilities than carrying out end-to-end cyberattacks. Based on its evaluation and the safeguards built into the model, the company believes GPT-5.6 Sol does not reach the "critical" capability threshold defined in its preparedness framework.The company said it believes the model will provide real benefits for legitimate defensive cybersecurity work, while still blocking improper offensive uses with its existing safeguards.The rollout comes as other top AI developers are also navigating similar government discussions. Anthropic also said it is working with the government on safeguards for its latest models. That suggests that oversight of frontier AI is becoming a broader industry issue, rather than one focused on a single company.What happens next?OpenAI says its immediate priority is expanding GPT-5.6 beyond the initial preview group while continuing discussions with the US government. The company expects broader availability in the coming weeks if no significant issues emerge during the ongoing evaluation period.The launch also coincides with broader regulatory developments in the US. As part of an Executive Order, the administration is expected to establish a classified process by August for assessing the cyber capabilities of advanced AI systems. That framework will determine which models qualify as "covered frontier models" requiring additional government oversight before release.For OpenAI, GPT-5.6 represents both a technical milestone and a test case for how future frontier AI models may be introduced under evolving government review mechanisms.Frequently Asked Questions1. What are the three models included in the GPT-5.6 family?OpenAI has introduced GPT-5.6 Sol, GPT-5.6 Terra and GPT-5.6 Luna. Sol is the flagship model, Terra balances performance and efficiency, while Luna focuses on speed and affordability.2. What is GPT-5.6's Ultra mode?According to OpenAI, Ultra mode allows the model to split work among multiple sub-agents to support more advanced reasoning tasks.3. Why is GPT-5.6 initially available to only a small number of companies?The company says the limited preview follows a US government review process, with participation currently restricted to around 20 approved organisations.4. What has OpenAI said about GPT-5.6's cybersecurity capabilities?OpenAI says GPT-5.6 Sol is better at helping users identify and fix vulnerabilities than conducting end-to-end attacks and does not reach the "critical" capability level in its preparedness framework.5. What happens after the current preview phase?OpenAI expects to expand access to more companies next week and aims for a broader rollout in the coming weeks, subject to the ongoing evaluation process.end of article