OpenAI has launched a restricted early preview of its latest generation of AI models, collectively known as GPT-5.6, including three variants named Sol, Terra, and Luna. The company is making the models available first to a small group of trusted partners approved by the U.S. government before wider distribution. The release marks another step in the rapid advancement of frontier AI systems, but it also highlights growing regulatory scrutiny over powerful new models. OpenAI previewed the models’ capabilities to government officials prior to the announcement and agreed to limit initial access at the administration’s request.
Model tiers target different use cases
Sol serves as the flagship model, designed for the most demanding reasoning and agentic tasks. Terra offers a balanced option positioned for everyday high-volume work at lower cost than the top tier. Luna provides faster and more affordable performance for lighter workloads. According to OpenAI, the new family delivers meaningful improvements in areas such as coding, scientific analysis, and cybersecurity tasks compared with previous versions. The company highlighted strong results on benchmarks like Terminal-Bench 2.1 for complex command-line workflows.
The rollout also comes shortly after OpenAI strengthened its financial position by securing an additional $10 billion in funding, providing greater resources to expand AI infrastructure and accelerate the development of future frontier models. This investment is expected to support the company’s long-term AI roadmap as competition in the sector continues to intensify.
Government coordination shapes rollout
OpenAI stated that it previewed its plans and model capabilities to the U.S. government as part of ongoing engagement. The initial limited preview is restricted to select trusted partners whose participation has been shared with authorities. The company plans to expand availability in the coming weeks.In its announcement, OpenAI noted it does not view this type of government access process as a desirable long-term standard. Officials cited national security and cybersecurity considerations in requesting the phased approach. The decision aligns with a recent executive order focused on evaluating advanced AI systems before public deployment.
The phased rollout also reflects a broader shift across the AI industry, where governments are taking a more active role in reviewing advanced AI systems before public release. Similar regulatory challenges have recently affected Anthropic, with its latest Claude models facing restrictions tied to U.S. export control policies, highlighting the growing intersection of AI innovation and national security.
Focus on safety and safeguards
The models ship with OpenAI’s most robust safety measures to date. These include strengthened protections against misuse in sensitive areas such as offensive cyber operations. Testing showed the models can assist effectively with vulnerability research and defensive security work while stopping short of enabling fully autonomous end-to-end attacks on hardened targets in evaluations.OpenAI emphasized that GPT-5.6 Sol remains below the “Cyber Critical” threshold in its internal preparedness framework. The company continues red-teaming and real-time monitoring during the preview phase.Pricing and availabilityDuring the limited preview, access is available through the OpenAI API and Codex platform to approved partners. Standard pricing starts at $5 per million input tokens and $30 per million output tokens for Sol, with lower rates for Terra and Luna. Broader rollout to ChatGPT users and general API customers is expected soon.This controlled launch reflects the evolving balance between innovation speed and risk management in the AI sector. As models grow more capable, coordination between leading developers and policymakers is becoming standard practice, even as companies push for predictable and innovation-friendly frameworks.
