OpenAI to Roll Out Advanced GPT-5.6 Models Sol, Terra and Luna on Thursday After Government Review

OpenAI is poised to broaden access to its latest family of artificial intelligence models, GPT-5.6, including the flagship Sol and more affordable variants Terra and Luna, on Thursday following a limited preview and additional national security evaluations requested by the U.S. government.

The rollout marks a significant step in the rapid evolution of frontier AI systems, coming amid heightened tensions in the U.S.-China technology competition and concerns over potential misuse of powerful models in cybersecurity and other sensitive domains.

OpenAI initially previewed the GPT-5.6 series on June 26 in a restricted manner for a small group of vetted partners, at the request of the Trump administration. Axios first reported the planned broader launch, noting that the administration approved it after further testing and consultations.

The company described Sol as its new flagship model, a "step function better" than GPT-5.5, optimized for complex reasoning, long-horizon agentic workflows, coding, scientific applications and cybersecurity tasks. Terra offers performance competitive with GPT-5.5 at roughly half the cost, while Luna prioritizes speed and affordability for high-volume work.

"Sol is our new flagship and a step function better than GPT-5.5. Terra delivers performance competitive to GPT-5.5 at 2x lower cost. Luna is our most cost-efficient model, delivering strong capability at our lowest cost," OpenAI stated in its announcement.

The models introduce new capabilities such as "max" reasoning effort for deeper analysis on Sol and an "ultra" mode that delegates tasks to multiple sub-agents for tackling intricate problems. GPT-5.6 Sol reportedly sets a new state of the art on Terminal-Bench 2.1, a benchmark evaluating complex command-line coding workflows involving planning, iteration and tool use.

Pricing through the API reflects the tiered approach: Sol 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. OpenAI also plans to make Sol available on Cerebras hardware starting in July, promising inference speeds of up to 750 tokens per second for select customers.

The delayed and staged release reflects growing U.S. government scrutiny of advanced AI. President Donald Trump signed an executive order establishing a voluntary framework for developers to submit "covered frontier models" for up to 30 days of government review before wider distribution to trusted partners. OpenAI said it previewed the models' capabilities to officials and agreed to the initial limited access.

"We believe in broad access and plan to make GPT-5.6 Sol, Terra and Luna generally available in the coming weeks. For now, at the request of the U.S. government, we're starting with a limited preview among a small group of trusted partners in Codex and the API," the company noted.

This approach mirrors actions taken with rival Anthropic, which temporarily restricted its advanced models Mythos 5 and Fable 5 following government export control concerns before lifting some curbs after implementing safeguards.

National security experts have warned that highly capable AI could accelerate sophisticated cyberattacks on critical infrastructure, which often relies on legacy systems. Both the U.S. and China are investing heavily in AI leadership, with Beijing also reportedly discussing restrictions on overseas access to its top models.

OpenAI emphasized improvements to its safety measures, describing GPT-5.6 Sol as launching with its "most robust safety stack yet." The company conducted extensive red teaming and automated testing, strengthening protections against high-risk cyber activity and misuse while aiming to support defensive cybersecurity work.

Under OpenAI's Preparedness Framework, all three models in the family were classified as high capability for cybersecurity and biological/chemical risks. Independent evaluations have highlighted strong performance in areas like vulnerability research, though the company acknowledged challenges in fully eliminating risks such as jailbreaks.

The launch comes as competition intensifies. Billionaire Elon Musk, whose xAI is developing rival systems, announced that Grok 4.5 — based on a new 1.5-trillion-parameter foundation model — is in private beta at SpaceX and Tesla, with plans for monthly new foundation model releases.

"Grok 4.5, based on our 1.5T V9 foundation model, with Cursor data added in supplemental training, is now in private beta at SpaceX & Tesla. Early evals show performance close to, perhaps exceeding Opus," Musk posted on X.

Industry analysts view the GPT-5.6 family as continuing OpenAI's push toward more efficient, specialized tiers rather than a single monolithic model. The celestial naming — Sol (Sun), Terra (Earth), Luna (Moon) — signals a durable capability structure where tiers can advance independently.

For developers and enterprises, the models promise better handling of agentic tasks, improved prompt caching for cost predictability, and stronger performance in biology, coding and quantitative reasoning. However, full public availability via ChatGPT and broader API access is still ramping up following the government-coordinated phase.

The developments underscore the balancing act between innovation speed and security in the AI sector. While companies like OpenAI stress the benefits for legitimate users — from researchers to cybersecurity defenders — regulators are focused on mitigating dual-use risks that could benefit adversaries.

As the Thursday rollout proceeds, observers will watch closely for how quickly access expands and whether the voluntary government review process sets a precedent for future releases. OpenAI has indicated it does not see the current limited approach as a long-term default.

The AI race shows no signs of slowing, with implications stretching across national security, economic competitiveness and technological progress. Thursday's launch of GPT-5.6 could provide a fresh benchmark for where the leading labs stand in this high-stakes field.