OpenAI has announced a restricted preview launch of its new GPT-5.6 model family — comprising Sol, Terra, and Luna — making them available initially only to trusted partners, following a request from the US government to delay a full public rollout. The company briefed federal authorities on GPT-5.6's capabilities ahead of launch and agreed to the limited release while regulators develop a broader evaluation framework under a recent cybersecurity executive order, according to multiple reports.

What the GPT-5.6 Models Offer

Sol is positioned as OpenAI's flagship model for demanding scientific workloads. It supports Max Reasoning Effort, a capability that allows the model to spend significantly more time working through complex problems before generating a response, and an Ultra Mode that deploys multiple specialised sub-agents working in parallel on coding, research, and multi-step tasks.

Terra is designed as a balanced option for enterprise and developer use cases, reportedly delivering performance comparable to GPT-5.5 at roughly half the cost. Luna, the third model, is built for speed and cost efficiency, targeting applications where affordability matters without compromising core capabilities.

Rollout Timeline and Hardware

OpenAI said it plans to make all three models available across ChatGPT, its API, and Codex in the coming weeks. Sol will be accessible on Cerebras hardware from July for select customers, with inference speeds of up to 750 tokens per second.

The company was explicit that it does not want government preview requirements to become a permanent standard for frontier AI releases — a signal that it views the current arrangement as a one-time accommodation rather than a policy precedent.

Why the US Government Intervened

The limited launch comes amid heightened scrutiny of frontier AI models in the United States. Federal authorities are working to establish a structured evaluation process for advanced AI systems, particularly those with capabilities in reasoning, coding, and cybersecurity — areas where GPT-5.6 reportedly advances the state of the art.

This is not an isolated instance. Separately, Anthropic recently received US approval to restore limited access to its Mythos 5 model for certain trusted partners, after resolving national security concerns. A government order issued two weeks prior had abruptly restricted foreign national access to Mythos 5 and a related model, Fable 5, over fears that security guardrails could be circumvented. The two cases together suggest a pattern of active federal oversight of high-capability AI deployments.

Broader Context

The GPT-5.6 launch reflects a growing tension between the pace of frontier AI development and the capacity of regulatory frameworks to keep up. OpenAI's decision to brief the government pre-launch and accept a staged rollout marks a notable shift in how leading AI labs are navigating Washington's increasing interest in advanced model governance. Whether this model of voluntary pre-clearance becomes industry standard remains to be seen.