OpenAI releases the GPT-5.6 series models, limited availability per the US government’s request, stating this should not become a long-term default practice.

OpenAI releases the GPT-5.6 series models, limited availability per the US government’s request, stating this should not become a long-term default practice.

```

OpenAI has officially released the GPT-5.6 series of models, including the flagship Sol, the balanced Terra, and the lightweight Luna. Unlike previous public launches, this time OpenAI has not immediately opened access across the board; instead, at the request of the U.S. government, preview access is only provided to a few “trusted partners.”

OpenAI stated that this is to cooperate with the U.S. government in establishing a security assessment process for frontier AI models, but at the same time, it unusually emphasized publicly that a government approval-based model release mechanism “should not become the long-term default mode.” This statement suggests that, against the backdrop of the Trump administration’s increasingly stringent national security regulations on advanced AI, the release of AI models by U.S. companies has entered a new phase: not only are model capabilities a focus of competition, but the release process itself is now subject to government security review.

GPT-5.6 Series: Positioning of Sol, Terra, and Luna

On Friday the 26th, Eastern Time, OpenAI unveiled three models in the GPT-5.6 series: “Our next-generation frontier model GPT-5.6 Sol; the balanced Terra model for both efficiency and everyday work; and the fast and affordable GPT-5.6 Luna, suitable for high-frequency, large-scale tasks.”

  • GPT-5.6 Sol: Positioned as the flagship model, targeted at the most complex reasoning, scientific research, software development, cybersecurity, biotechnology, and agent workflows.
  • GPT-5.6 Terra: Focuses on comprehensive capabilities, maintains similar performance to GPT-5.5 but at roughly 50% lower cost, targeted at enterprises and everyday productivity scenarios.
  • GPT-5.6 Luna: Fastest and cheapest, suitable for large-scale online services and high-throughput scenarios.

Among them, Sol adds advanced modes, including:

  • Max: Longer thinking time and deeper reasoning;
  • Ultra: Capable of coordinating multiple sub-Agents to complete complex, long-term tasks.

OpenAI reports that the new models have made significant improvements in software engineering, autonomous agents, cybersecurity, and defense research, undergoing over 700,000 GPU hours of automated safety testing and extensive external red teaming.

In terms of pricing, GPT-5.6 Sol charges $5 per million input tokens and $30 per million output tokens, which is about half the cost of Anthropic’s Claude Fable 5 model (which is $10/$50 per million input/output tokens, respectively). Terra’s price is half of Sol’s, and Luna’s is less than half of Terra’s.

Biggest Change Is in Release Method: U.S. Government Requires Restricted Access

Compared to model performance, the market is paying more attention to this release method.

OpenAI announced that, for now, GPT-5.6 is only open for preview to a limited number of trusted partners. This arrangement was not OpenAI’s own initiative, but done per the Trump administration’s request. OpenAI disclosed in its announcement:

“As part of our ongoing communication with the U.S. government, before today's release, we provided a preview of our plans and the model’s capabilities. At the government’s request, we will first provide a limited preview to a small group of trusted partners—whose involvement has been reported to the government—then expand to a broader release.”

Media outlets pointed out that the U.S. government wants to establish a unified review process for frontier models with potential cybersecurity capabilities before public release, thus requiring OpenAI to adopt a staged access approach granting first access only to approved organizations. The first batch of approved companies is about twenty, with broader access planned in the next few weeks.

Reports say that under the current mechanism, some customers’ model access even needs individual government approval.

This arrangement follows the Trump administration’s recent push for stricter regulations on advanced AI models. Previously, the U.S. government had required Anthropic to retract the open release of some new models and to establish a review framework for "Covered Frontier Models".

OpenAI Expresses Concern About Government Approval, Hopes for a Repeatable Release Process

Although cooperating with U.S. government requirements to limit access, OpenAI’s official announcement included a notably strong statement.

The announcement starts by stating that OpenAI “firmly believes there should be broad access, and plans to fully roll out GPT-5.6 Sol, Terra, and Luna models in the coming weeks.” It also says that during the preview period, they’ll continue testing and work closely with partners to facilitate greater access, then writes:

“We believe this kind of government-mediated access process should not become the long-term default mode. It blocks users, developers, businesses, cybersecurity defenders and global partners who truly need these top-tier tools from accessing them.

We are taking this temporary measure because we believe it is the best way to ensure broader release in the coming weeks; meanwhile, we will work with government to develop an executive order framework for cybersecurity and establish a repeatable process for future model releases.”

This is one of the most noteworthy statements in OpenAI’s announcement.

The company, on one hand, clearly supports the Trump administration’s goal of safeguarding national security; on the other, it emphasizes that advanced AI models should be quickly opened to developers, research institutions, enterprises, and cybersecurity defenders, as all these groups need timely access to advanced AI capabilities.

OpenAI further stated that it hopes this preview period can help it and the government jointly establish a clear, repeatable approval process, instead of making ad-hoc government approval the de facto long-term norm for the AI industry.

Model Safety Evaluation: Not Yet at the Highest Risk Level

OpenAI also released its latest deployment safety evaluation results.

According to the Deployment Safety report, although GPT-5.6 has further enhanced autonomous execution, vulnerability analysis, and complex reasoning capabilities, the company believes it has not yet reached the danger threshold requiring the highest level of restriction in the Preparedness Framework.

OpenAI stated: the model’s cybersecurity capabilities are significantly improved; it is better at finding software vulnerabilities than assisting with attacks; it has not reached a danger level requiring full deployment restrictions; it was evaluated by automated testing, expert red teaming, and external institutions before release.

Codex, ChatGPT Will Also Be Upgraded to GPT-5.6

Besides the API, OpenAI stated:

  • ChatGPT will gradually integrate the GPT-5.6 series in the future;
  • Codex will also be upgraded to the GPT-5.6 model;
  • The developer platform will then open up wider access.

Media reports point out that this upgrade means that OpenAI’s entire product line—including ChatGPT, Codex, and the API—will gradually transition to the GPT-5.6 architecture, though the timing of full access still depends on the progress of government review.

AI Competition Enters a New Phase: Competing Not Just on Models, But Also on Regulation

Industry insiders believe the greatest significance of the GPT-5.6 release is not just further improvements in model performance, but the first deep intervention by the U.S. government in the release of the world’s most advanced AI models.

Following Anthropic, OpenAI also has to undergo government security reviews before model launch, meaning U.S. AI regulation is shifting from traditional industry oversight to frontier model deployment management.

However, OpenAI’s public emphasis that government approval should not be the long-term default shows the company hopes the current arrangement is only a temporary transition, not a future long-term rule for the industry.

In the coming weeks, as GPT-5.6's access widens and the U.S. government formally establishes frontier model review procedures, how the AI industry balances national security with technological openness will become the new focus of global AI competition.

Risk Warning and DisclaimerThe market involves risks, and investment should be cautious. This article does not constitute personal investment advice and does not take into account the individual user's specific investment objectives, financial situation, or needs. Users should consider whether any opinions, views, or conclusions in this article are suitable for their circumstances. Investing based on this information is at your own risk. ```