OpenAI has announced a restricted release for its latest suite of AI models, making them available only to a "small group of trusted partners" following a directive from the U.S. government.
The new GPT-5.6 generation comprises Sol, the company's most powerful flagship model; Terra, designed as a more balanced option for general use; and Luna, a faster, more cost-effective alternative. Despite Sol's advanced capabilities, the Trump administration has imposed restrictions on the deployment of all three models. OpenAI clarified that this preview is exclusively for partners "whose participation has been shared with the government."
This request from the administration aligns with broader efforts by the U.S. government to exert greater control over advanced AI systems developed by technology companies. A notable precedent occurred when Anthropic launched its powerful public model, Fable 5. The administration subsequently mandated that Anthropic revoke access for all foreign nationals, prompting the company to withdraw the model entirely.
The incident has ignited discussions regarding the appropriate extent of governmental authority over the release of AI models. Dean Ball, a former White House AI advisor and an impending OpenAI employee, suggests that President Trump's recent executive order—which encourages select AI companies to voluntarily submit their most advanced models for government review up to 30 days prior to release—has effectively established a "de facto involuntary licensing regime" for cutting-edge AI, leading to considerable restrictions.
Ball further contends that this issue is exacerbated by the government's lack of clearly defined safety standards. Such ambiguity could result in indefinite launch delays, potentially granting China a strategic advantage in the global AI race and jeopardizing the billions of dollars invested in AI infrastructure development.
While OpenAI complied with the administration's request on this occasion, the AI firm explicitly conveyed its dissatisfaction with the arrangement.
"We don’t believe this kind of government access process should become the long-term default," stated a blog post released by OpenAI on Friday. The company argued, "It keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them."
OpenAI characterized the current preview as a "short-term step" intended to pave the way for broader GPT-5.6 availability in the coming weeks. The company is actively collaborating with the administration to formulate a new executive order framework on cybersecurity and establish a "repeatable process for future model releases."
OpenAI highlights GPT-5.6 Sol as its most robust model to date, featuring enhanced agentic capabilities across coding, biology, and cybersecurity. Sol introduces a "max" reasoning effort mode and an "ultra" mode, which utilizes coordinated subagents to tackle highly complex tasks—a feature that, while powerful, significantly increases token usage.
GPT-5.6 demonstrates strong performance across various benchmarks, according to OpenAI. It reportedly surpasses Anthropic’s Claude Mythos 5 in coding workflows, a model that the Trump administration also effectively restricted this month. OpenAI further notes that GPT-5.6 Sol is competitive with the Mythos preview while consuming only a third of the output tokens.
To mitigate concerns about the safety of its powerful models, OpenAI states that Sol incorporates its most advanced security stack. The company emphasizes that Sol is heavily fortified against adversarial attacks and has been intentionally optimized to prioritize defensive cybersecurity work over offensive exploits. This means it is designed to be highly resistant to jailbreaking and focuses on educating users about defensive strategies rather than facilitating system compromises.
OpenAI also asserts that its safety guardrails are intrinsically embedded within the core model's behavior, rather than relying on an external filter. This approach likely aims to circumvent issues similar to those encountered by Anthropic with Fable 5. During its brief public availability, Fable 5's classifiers, upon detecting high-risk topics such as cybersecurity, biology, or chemistry, would not merely block the prompt but instead reroute the request to an older model. This overly cautious and invisible downrouting mechanism led to numerous false positives and considerable user frustration.
Although the GPT-5.6 models are initially restricted to a select group of partners, OpenAI intends to make them more widely accessible to users of ChatGPT, Codex, and the API in the near future.
GPT-5.6 will be offered in three sizes with tiered pricing: Sol is priced at $5 per million input tokens and $30 per million output tokens; Terra costs half of Sol's rates; and Luna is available for $1 per million input tokens and $6 per million output tokens. OpenAI also notes improvements in prompt caching, designed to make repeated prompts more economical and predictable.
The Editorial Staff at AIChief is a team of professional content writers with extensive experience in AI and marketing. Founded in 2025, AIChief has quickly grown into the largest free AI resource hub in the industry.