On July 9, 2026, OpenAI publicly launched GPT-5.6, releasing three distinct variants—Sol, Terra, and Luna—to the market. While this rollout provides broad access, it simultaneously codifies a permanent divide in the AI ecosystem, marking the transition of the tiered model economy from policy proposal to commercial reality.
This launch serves as the second half of our investigation into the new infrastructure of power. In our previous report, The Federal Gate, we detailed the secret preview period where access to the flagship Sol model was restricted to approximately 20 government-vetted organizations at the request of the White House ONCD and OSTP. Today, the public can access the lower tiers, but the most potent capabilities remain locked behind a federal wall.
Terra, which offers performance comparable to GPT-5.5 at roughly half the cost, and the budget-friendly Luna are now available to the general public. However, Sol — the flagship model capable of finding vulnerabilities and pieces of exploits — is subject to a permanent restriction. Even with general availability, OpenAI has confirmed that the most sensitive cybersecurity and biological capabilities within Sol will remain gated for trusted defenders.
The mechanism behind this power is physical. Sol runs on Cerebras wafer-scale silicon, delivering approximately 750 tokens per second in weight-streaming mode, supported by a massive 750-megawatt infrastructure deal. This is not just software; it is a capital-intensive, state-aligned operation.
This architecture forces a permanent divergence in the developer ecosystem. By restricting the full suite of Sol’s capabilities to a select group, the government and OpenAI have institutionalized a two-class system for agent building. Approved organizations operate with a toolset that the rest of the market is structurally barred from reaching. The most powerful tools are no longer just for sale — they are for the cleared.