OpenAI releases GPT-5.6 Sol to the public after delays imposed by the White House

OpenAI has begun the public rollout of GPT-5.6 Sol, which the company itself describes as its "most powerful model to date." The announcement was made on Wednesday via X, and the gradual rollout started on July 9. Alongside Sol, OpenAI is introducing two lighter models: Terra, conceived as an "everyday" option, and Luna, a lower-cost version. It is a tiered family that follows the industry's usual pattern: a flagship model with maximum capabilities accompanied by cheaper variants for different use cases.

The most striking part of the news is not so much the model itself as the regulatory context surrounding its release. According to the article, the White House had asked OpenAI to limit the initial launch of GPT-5.6 to a small evaluation group. In its own blog, OpenAI described that group as "a small set of trusted partners whose participation has been shared with the government." In other words, before opening the model to the general public, Washington wanted visibility—and apparently veto power—over who tested the system and under what conditions. Now, according to Mashable, OpenAI already has the green light for general release.

In that same blog, the company explains that it spent weeks stress-testing GPT-5.6 Sol for weaknesses, a process that led to "strengthening protections for high-risk activities, sensitive cybersecurity-related requests and repeated misuse." OpenAI presents the model as "the most capable so far in cybersecurity," and also makes a specific comparative claim: it maintains that GPT-5.6 Sol matches the performance of Claude Mythos 5, from its competitor Anthropic, using only a third of the output tokens (that is, the blocks of text the model processes to generate its responses). It is an efficiency claim—doing the same with less compute—although the article provides no benchmarks or methodology behind that comparison, so it should be taken as a promotional statement from OpenAI itself, not as an independently verified figure.

The episode of government intervention is neither isolated nor exclusive to OpenAI. The article notes that Anthropic faced similar restrictions: the White House ordered the company to "suspend all access to Fable 5 and Mythos 5 for any foreign national, whether inside or outside the United States." Anthropic subsequently received permission to share Mythos 5 with "a set of U.S. organizations that operate and defend critical infrastructure," while Fable 5—described as a version of Mythos—is already available globally. This paints a picture in which the U.S. administration is exercising active and differentiated control over which frontier models can leave the country and on what terms, treating the major AI labs in a manner similar to how the export of dual-use or weapons technology has historically been treated.

The article connects this regulatory control with another recent development: a week before this announcement, OpenAI reportedly proposed handing the U.S. government a five percent stake in the company, with CEO Sam Altman holding preliminary talks with President Donald Trump, as reported by the Financial Times. Although Mashable does not go into the details of that negotiation, the juxtaposition of the two stories suggests an increasingly intertwined relationship between OpenAI and the White House, in which access to releasing frontier models appears to go hand in hand with some kind of closer government agreement or oversight.

Overall, the piece is more a chronicle of AI governance than a technical analysis of GPT-5.6 Sol: there are hardly any details about architecture, specific capabilities beyond cybersecurity, or pricing. What is relevant for anyone following the agentic AI race is the precedent being set: frontier model launches by the two leading U.S. companies no longer depend solely on each lab's internal safety criteria, but are subject to conditions imposed directly by the White House, with different nuances depending on the model and the user's destination country.