OpenAI's New AI Was So Powerful, It Had To Brief The US Government Before Release

OpenAI has announced a preview of GPT-5.6 Sol, a new artificial intelligence model that the company says represents its most capable system to date. The announcement comes as governments and technology companies continue to debate how powerful AI systems should be developed and released.

According to a blog post published by OpenAI, GPT-5.6 Sol has been designed to perform better in areas such as computer programming, scientific research, cybersecurity and long-term problem-solving. The company says the model can handle more complex tasks and provide more reliable responses than earlier versions.

The new system is part of a broader GPT-5.6 family of models. OpenAI has also introduced two additional versions called Terra and Luna, which are intended for different levels of performance and cost.

What Can It Do?

GPT-5.6 Sol is built for complex, long-running tasks, featuring "max" reasoning for deeper analysis and an "ultra" mode using multiple sub-agents. It sets a new coding benchmark on Terminal-Bench 2.1 and shows advances in biology and genomics research. In cybersecurity, it is OpenAI's most capable model for extended tasks like vulnerability detection and exploitation.

Shift from hands-off to tighter AI controls under Trump administration

According to The Washington Post, the US federal government will now vet companies seeking access to OpenAI's latest model, GPT-5.6, marking a significant expansion of regulation in the AI sector. 

OpenAI said in a Friday blog post that access approvals will initially be managed by the government as longer-term rules are developed in coordination with the administration. The move follows a recent decision to block rival Anthropic from offering its most advanced models to non-US users. 

These actions signal a major policy shift under President Donald Trump, who had earlier favored minimal regulation but has reconsidered after concerns grew over powerful AI systems capable of identifying software vulnerabilities. The evolving approach shows rising alarm in Washington and globally over the potential risks posed by advanced artificial intelligence technologies.

Who Can Use It and How Much Does It Cost?

During the preview, GPT-5.6 models will initially be available through the API and Codex to a select group of trusted partners, with broader availability planned for ChatGPT users and developers soon. Sol is priced at $5 per million input tokens and $30 per million output tokens, Terra at $2.50 input and $15 output, and Luna at $1 input and $6 output.