OpenAI unveils GPT-5.6 in the middle of the US AI game

Less than 24 hours after the news broke that OpenAI would do so stagger its next model release At the request of the Trump administration, that model, GPT-5.6, is here. On Friday, the company revealed a limited preview of its new version of GPT 5.6: Sol, the flagship; Terra, a medium-sized model of “high performance”; and Luna, a “fast and affordable” daily. OpenAI claims to specialize in coding, cybersecurity, and biology, as well as being focused on long-term AI projects.

For a million tokens, GPT-5.6 Sol is priced at $5/$30 output (about half the price of Anthropic’s Claude Fable 5, which is $10 input/$50 output). Terra is half the value of Sol, and Luna is less than half the value of Terra. The company also released two additional versions of Sol: the “max” mode for serious thinking and the “ultra” support for assistants – the release of OpenClaw, and possibly the trademark of the OpenClaw developer. Peter Steinbergerwork at OpenAI so far.

Unsurprisingly among a security scare in Washington, DCOpenAI has dedicated its blog post information to security and abuse. It seemed to fit in with recent jailbreaking problems of its competitor Anthropic, writes that “GPT-5.6 is trained to resist illegal cyber attacks, including when users try to hide their data or jailbreak.” It also said that Sol’s version “is better for helping people find and fix vulnerabilities than it is to reliably attack criminals,” and that Sol does not cross cyber-critical boundaries according to the OpenAI design process – although it should be noted that OpenAI. recently updated His plan for April is to remove some areas of the previous lesson.

The company said Sol has “the strongest security to date” and that it “enhances security for high-risk applications, cyber requests, and repeated exploits.” OpenAI said it dedicated “nearly 700,000 hours of A100e GPU” to the red machine and also worked with third-party testers, the latter of which will continue testing for the next two weeks.

OpenAI also seems to be taking a more proactive approach during the preview, which is being watched by Trump. The company wrote that “in some cases defenses can intervene in legitimate activities, especially in dual-use areas where defenses and offensives can appear to be the same. The report earlier this week said that the Trump administration will approve customers in any case during the observation period.

OpenAI said that the prototype team should be available in the coming weeks because the company believes that it is a “broad approach,” and that the company has agreed with the US government before this, but hopefully it will not become the norm.

“We do not believe that such government acquisition measures should be temporary,” the company wrote. “It preserves the best tools from users, developers, businesses, cyber defenders, and partners around the world who need them. We are taking this short step because we believe it is the most powerful way to achieve greater availability in the coming weeks, while we work with the Ministry to develop a cyber Executive Order and an iterative process to release future models.”