JAKARTA — GPT-5.6 Sol, OpenAI’s newest and most powerful AI model, is not going public right away. The company is rolling it out in phases after a request from the U.S. government, with early access limited to customers who have already been approved by the government.
The move marks a meaningful shift in how advanced AI models are launched. This is not just about new features. For the industry, it shows that next-generation AI now sits squarely in the national security lane.
GPT-5.6 Sol, OpenAI’s strongest model
OpenAI describes GPT-5.6 Sol as the flagship model in a new family that also includes Terra, a more efficient version, and Luna, a lower-cost model. In its explanation, the company said Sol can complete about 50 percent of long-horizon professional tasks and posted its highest score among OpenAI’s earlier models for coding ability.
At this point, OpenAI is not only chasing speed. Sol is also built for work that needs extended reasoning. The company has prepared two new modes, “max” and “ultra,” which give the model more time to think and coordinate agents for specific tasks.
OpenAI says all three models will become more widely available in the coming weeks. But for the first stage, the door is narrow. Very narrow.
Why access is being limited by the U.S. government
The Information first reported that the Trump administration asked OpenAI to delay a full launch for security reasons. OpenAI then confirmed it was opening a limited preview to a small group of trusted partners, with the names shared with the U.S. government.
The company says the step is part of a voluntary process tied to an existing collaboration with the government. OpenAI also stressed that it does not want this kind of access model to become a long-term standard.
“We are taking this short-term step because we believe it is the strongest path to broader availability in the coming weeks,” OpenAI wrote in its official blog. The company added that it is working with the Trump administration to build an executive-order framework for cybersecurity and a repeatable process for future model launches.
Cybersecurity is the most sensitive area
OpenAI highlighted one area that worries regulators the most: cybersecurity. Sol is said to be strong at finding weaknesses and exploiting vulnerabilities. In some tests, OpenAI previously said the model was on par with Anthropic’s cyber model Mythos, although it used far fewer tokens.
According to OpenAI, GPT-5.6 Sol used about one-third of the tokens consumed by Mythos, but it still trailed Anthropic’s Mythos 5, which is viewed as stronger. Differences this small matter, because in frontier models, tiny gaps can translate into major real-world consequences.
OpenAI also said Sol has not yet crossed its internal “Cyber Critical” threshold. When tested with Firefox and Chrome, the model found the beginnings of an exploit, but did not produce a working exploit. To look for other weaknesses, OpenAI said it had already spent 700,000 GPU hours testing the model’s security. Two more weeks of human testing are still due before a wider release.
A new precedent in Washington and what it means for AI
The restriction is not happening in isolation. Over the past month, frontier AI labs have faced tighter access controls more than once. In early June, the U.S. Commerce Department imposed export controls on Anthropic and forced the company to cut off overseas access to two of its top models, citing national security concerns. Anthropic objected, but still had to pull the models from public access.
Trump also signed an executive order earlier this month directing federal agencies to create a framework that would let AI companies give the government early access up to 30 days before a broader release. OpenAI says what it is doing now is still voluntary, unlike Anthropic’s case, where the company was compelled.
The problem is that the framework is not fully in place yet. That is one reason OpenAI chose a phased launch. From the industry’s view, this could be the first test of a filtering system for advanced AI models. From the public’s view, it is a sign that access to the most powerful models may no longer be automatic.
Pricing, early users, and what it means for readers
OpenAI has also set premium pricing for GPT-5.6 Sol: $5 per one million input tokens and $30 per one million output tokens. Terra is priced at $2.50 and $15, while Luna is much cheaper at $1 and $6. The price gap shows that Sol is aimed at heavy-duty use, not just everyday chat.
Early users are customers already approved by the U.S. government. OpenAI said that list will be expanded next week. The mechanism, according to the company, works like this: share customer names with the government, then receive feedback before access opens wider.
For readers outside the United States, the story matters because Washington’s policy direction often shapes the global market. If frontier models start being treated like sensitive technology, other AI companies may face similar scrutiny. Regular users may not feel it today. But pricing, access, and feature limits could change down the line.
OpenAI itself is walking a careful line. It wants wider access, but it also does not want its most powerful model to fall into the wrong hands. Between those two goals, Sol is being launched with the brakes on.
And one number keeps hanging over the rollout: 700,000 GPU hours. That is how much compute OpenAI says it spent testing the model’s security before letting it near the public.
Brief summary
1. GPT-5.6 Sol is being launched in phases after a U.S. government request, not opened to everyone at once.
2. OpenAI is highlighting coding and cybersecurity strength while preparing “max” and “ultra” modes.
3. The rollout shows frontier AI models are increasingly being treated as a national security issue, not just a tech product.
Quick FAQ
What is GPT-5.6 Sol? OpenAI’s newest AI model and the strongest in its new lineup, alongside Terra and Luna.
Who gets early access? Customers approved by the U.S. government and trusted partners whose names OpenAI shared with the government.
Why is the launch limited? Because the U.S. government asked for a phased rollout over security concerns, especially around cyber capabilities.
📝 Leave a Comment
Comment as . Reviewed by an admin before it appears.