OpenAI will publicly release its GPT-5.6 Sol, Terra, and Luna models on Thursday, according to CNBC. The rollout follows a period of restricted access mandated by the U.S. government to assess model capabilities before a general public launch.
The Staged Release of GPT-5.6 Sol, Terra, and Luna
OpenAI announced the public expansion of its GPT-5.6 family late Tuesday. The release includes three distinct models tailored for different use cases. Business Insider reports that Sol serves as the flagship model, featuring enhanced agentic abilities in biology, cybersecurity, and coding. Terra is designed for everyday work, while Luna focuses on speed and affordability.
The company previously limited the rollout to a small group of trusted partners, a move CNBC notes was done at the request of the U.S. government. OpenAI’s internal blog stated that the participation of these select organizations was shared with federal authorities.
“Happy building,”
Sam Altman, CEO of OpenAI
Altman posted the message to X on Tuesday night as the company moved to expand preview access globally.
How the Trump Administration’s AI Executive Order Changed Deployment
The restriction on GPT-5.6 is part of a broader shift in how the U.S. government manages frontier AI. Under an executive order signed in June, President Donald Trump requested that developers voluntarily provide cutting-edge models to the government for capability assessments prior to full release.
Federal agencies were given 60 days to develop a formal evaluation process. OpenAI stated in June that it is collaborating with the government to create a repeatable process for future model releases and a framework for these assessments.
OpenAI has expressed reservations about this model of governance. The company stated at the time of the initial restricted release that it does not believe government access processes should become the long-term default, arguing that such delays keep essential tools from developers, enterprises, and cyber defenders.
The Anthropic Precedent: Jailbreaking and Export Controls
The rollout of GPT-5.6 follows a more volatile clash between the government and Anthropic. According to letsdatascience.com, the U.S. government exercised export control powers to force an 18-day suspension of Anthropic’s Claude Fable 5 and Mythos 5 models.
The timeline of the Anthropic intervention reveals a rapid escalation:
- June 9, 2026: Anthropic launches Claude Fable 5.
- June 12, 2026: Commerce Secretary Howard Lutnick directs Anthropic to suspend access for all foreign nationals.
- June 26, 2026: Commerce allows restored access for roughly 100 approved companies and agencies.
- July 1, 2026: Fable 5 controls are lifted for the general public.
“Our understanding is that the government believes it has become aware of a method of bypassing, or ‘jailbreaking’ Fable 5.”
Anthropic, via letsdatascience.com
Anthropic disagreed with the necessity of the recall, claiming the “jailbreak” simply involved asking the model to identify software flaws—a capability they noted was already available in other models, including OpenAI’s GPT-5.5.
Competitive Pressure: Grok 4.5 and Gemini 3.5 Pro
The timing of the GPT-5.6 release coincides with a surge of new frontier models. Hours after OpenAI’s announcement, Elon Musk announced that SpaceXAI would launch Grok 4.5 in the coming days. Musk described Grok 4.5 as an Opus-class model, but faster, more token-efficient and lower cost, directly referencing Anthropic’s high-end model family.
This clash continues the public rivalry between Musk and Sam Altman. Musk previously sued OpenAI in 2024, alleging the company betrayed its nonprofit origins, though a jury ruled against him earlier this year because he waited too long to file.
Beyond the Musk-Altman feud, Google is also entering the fray. Business Insider reports that Google plans to release Gemini 3.5 Pro later this month.
What’s Next for AI Governance and Benchmarks
The industry is now moving toward a standardized set of security rules. The Trump administration faces an August 2026 deadline to establish standardized benchmarks for evaluating the security risks of new AI models.
These benchmarks will likely determine what constitutes a disqualifying jailbreak and how future export control decisions are made. For enterprises, this means the government has proven it will pull commercial models from the market if they fail to meet undisclosed national security criteria.
The current trend across OpenAI, Anthropic, and xAI also shows a pivot toward efficiency. By introducing “lower cost” and “token-efficient” versions of flagship models, labs are responding to corporate customers who are increasingly conscious of their token spend.
Find more reporting in our Business section.
Worth a look