OpenAI announced on Friday that it is restricting the launch of its latest AI models to a "small group of trusted partners," following a request from the U.S. government. The next-generation GPT-5.6 lineup includes three models: Sol, the flagship model; Terra, a balanced option for everyday tasks; and Luna, a faster, lower-cost alternative. While Sol is the company's most powerful model, the Trump administration has limited the release of all three. OpenAI clarified that the preview is only available to partners "whose participation has been shared with the government."

This government request comes amid increasing pressure on AI companies to restrict their most advanced systems. After Anthropic released its powerful public model, Fable 5, the administration ordered the company to block access for any foreign nationals, leading Anthropic to remove the model entirely. This incident has raised questions about the extent of government authority over AI model releases. Dean Ball, a former White House AI advisor who will soon join OpenAI, argues that President Trump's recent executive order—which asks certain AI companies to voluntarily submit their most advanced models for government review up to 30 days before release—has effectively created an involuntary licensing system for frontier AI, resulting in heavy-handed restrictions. Ball contends that the problem worsens when the government lacks clearly defined safety standards, potentially causing endless launch delays that could not only give China an edge in the AI race but also jeopardize billions of dollars in AI infrastructure investments. Although OpenAI complied with the administration's request this time, the company made its displeasure clear. "We don't believe this kind of government access process should become the long-term default," stated a Friday blog post. "It keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them."

OpenAI described the preview as a "short-term step" that will pave the way for broader availability in the coming weeks, as the company collaborates with the administration to develop a new executive order framework on cybersecurity and a "repeatable process for future model releases."

**GPT-5.6 Sol Specs**

OpenAI claims that GPT-5.6 Sol is its strongest model yet, featuring enhanced agentic capabilities in coding, biology, and cybersecurity. Sol introduces a "max" reasoning effort mode and an "ultra" mode that employs coordinated subagents to tackle highly complex tasks—a sophisticated feature that significantly increases token usage. According to OpenAI, GPT-5.6 excels on several benchmarks, including slightly outperforming Anthropic's Claude Mythos 5 in coding workflows—a model that the Trump administration also effectively banned this month. OpenAI notes that GPT-5.6 Sol is competitive with the Mythos preview but uses only one-third of the output tokens. To address concerns about the safety of its powerful models, OpenAI says Sol incorporates its most robust security stack yet. The model is heavily hardened against adversarial attacks and intentionally optimized to favor defensive cybersecurity work over offensive exploits. In other words, it is designed to be difficult to jailbreak, prioritizing showing users how to defend against exploits rather than how to hack into systems. OpenAI also states that its safety guardrails are built directly into the core model's behavior, rather than relying on a separate filter on top of it. The company appears to be trying to avoid the pitfalls that ensnared Anthropic with Fable 5. During the brief period when Fable 5 was available, whenever the model's classifiers detected a high-risk topic—such as cybersecurity, biology, or chemistry—it wouldn't just block the prompt; it would route the request to an older model. This overly cautious flow and invisible downrouting led to numerous false positives and user backlash. While the GPT-5.6 models are initially limited to a select group of partners, OpenAI plans to make them more broadly accessible to users of ChatGPT, Codex, and the API soon. GPT-5.6 comes in three sizes with tiered pricing: Sol costs $5 per million input tokens and $30 per million output tokens; Terra costs half that; and Luna costs $1 and $6, respectively. OpenAI also says it has improved prompt caching to make repeated prompts cheaper and more predictable.