OpenAI announced on Friday that its latest AI models would be made available only to a “select group of trusted partners” at the request of the U.S. government.

The upcoming GPT-5.6 series features Sol, the company’s flagship model; Terra, a model optimized for general use; and Luna, a quicker and more budget-friendly alternative. Though Sol is the company’s most powerful model, the Trump administration has prohibited the launch of all three. OpenAI indicated that the preview is restricted to partners “whose involvement has been communicated to the government.”

This request from the administration arrives as the US government intensifies its demands on AI firms to limit their most sophisticated systems. Following the release of Anthropic’s most advanced public model Fable 5, the administration mandated that the firm restrict access for any foreign nationals, resulting in Anthropic taking the model offline completely. 

This situation has raised concerns about the extent of governmental authority over the release of AI models. Dean Ball, a previous AI advisor at the White House and soon-to-be OpenAI staff member, claims that President Trump’s recent executive order—requesting specific AI companies to voluntarily submit their most advanced models for government evaluation up to 30 days prior to their release—has established a de facto involuntary licensing system for frontier AI, resulting in excessive restrictions. 

Ball further argues that the issue worsens when there are no clearly defined safety standards from the government, which may lead to endless delays in launches that could not only benefit China in the AI competition but also threaten the significant investments in AI infrastructure. 

While OpenAI complied with the administration’s request this time, the company made it clear that it was dissatisfied with the situation.

“We believe this form of government access process should not become the long-term norm,” stated a blog post on Friday. “It restricts the best tools from users, developers, businesses, cybersecurity defenders, and international partners who need them.”

OpenAI described the preview as a “temporary measure” that will place GPT-5.6 on the trajectory toward wider availability in the coming weeks, as the company collaborates with the administration to create a new executive order framework on cybersecurity, along with a “repeatable process for future model releases.”

GPT-5.6 Sol specifications

OpenAI claims that GPT-5.6 Sol is its most powerful model to date, featuring enhanced capabilities in coding, biology, and cybersecurity. Sol introduces a “max” reasoning effort mode and an “ultra” mode that employs coordinated subagents to tackle highly intricate tasks (the kind of clever feature that significantly increases your token usage).

According to OpenAI, GPT-5.6 outperforms several benchmarks, including slightly surpassing Anthropic’s Claude Mythos 5 in coding workflows, which the Trump administration also effectively barred this month. OpenAI asserts that GPT-5.6 Sol is competitive with the Mythos preview while utilizing only a third of the output tokens. 

To alleviate concerns regarding the safety of its powerful models, OpenAI states that Sol is equipped with its most advanced security stack to date. OpenAI claims it is rigorously fortified against adversarial attacks and intentionally built to prioritize defensive cybersecurity tasks over offensive strategies. In essence, it’s engineered to be difficult to compromise while emphasizing user education on defense against exploits rather than hacking into systems. 

OpenAI also mentions that its safety guardrails are integrated into the core behavior of the model instead of relying on a separate filter. The company likely aims to evade the pitfalls that befell Anthropic with Fable 5. During the brief availability of Fable 5, its classifiers routed requests involving high-risk topics—such as cybersecurity, biology, or chemistry—not only blocked the prompts but redirected them to an older model. This overly cautious process and invisible downrouting resulted in numerous false positives and user dissatisfaction. 

Though the GPT-5.6 models are currently limited to a select handful of partners, OpenAI intends to expand access to users of ChatGPT, Codex, and the API shortly. 

GPT-5.6 is offered in three variations with tiered pricing: Sol is priced at $5 per million input tokens and $30 per million output tokens; Terra is half that price; and Luna costs $1 and $6, respectively. OpenAI also claims to have enhanced prompt caching to make repeated prompts more cost-effective and predictable.

OpenAI is once again making a significant, noteworthy commitment to India. The company has chosen former president of Uber India and South Asia, Prabhjeet Singh, to be its inaugural managing director in the country to enhance its presence in what it refers to as its second-largest market after the U.S.

Singh, who announced his departure from Uber on Friday, will begin his role at OpenAI in September and will report to Kiran Mani, the company’s managing director for Asia-Pacific, as communicated to TechCrunch. He will oversee OpenAI’s operations in India, focusing on consumer growth, enterprise engagement, partnerships, regulatory interactions, and overall operations, the organization stated.

This appointment signifies OpenAI’s latest commitment to India. The company launched its inaugural office in New Delhi last August and earlier this year indicated plans to open additional offices in Mumbai and Bengaluru. In 2024, it brought on former Truecaller and Meta executive Pragya Misra to spearhead public policy and partnerships before expanding her role to encompass strategy and global affairs last year. OpenAI had previously recruited former Twitter India head Rishi Jaitly as a senior advisor to assist with its collaboration with the Indian government regarding AI policy.

In recent months, OpenAI has formed partnerships in India that encompass higher education, enterprise financial services, AI-driven commerce, and web streaming, while also contributing to the country’s expanding data center infrastructure. OpenAI has cited India’s swiftly growing adoption of ChatGPT as an indicator of the market’s significance. Major Indian corporations like Reliance and Tata Group are also among its initial partners in the region.

The company has concurrently accelerated hiring in India, with positions available for AI deployment engineers, developer experience engineers, a developer marketing lead, a partner director, and solutions engineers.

India has become a crucial battleground for American AI enterprises, fueled by its extensive developer community, over a billion internet users, and increasing demand for generative AI. Competitor Anthropic inaugurated its India office in Bengaluru in late 2025 and appointed former Microsoft India managing director Irina Ghose as its head in India earlier this year.