OpenAI publicly launched GPT-5.6 this week, and it landed as a family rather than a single model. The lineup is Sol (the flagship, presumably massive), Terra (the mid-range workhorse), and Luna (the lightweight edge-friendly variant). But what makes this launch genuinely different from GPT-5, GPT-4.5, or any prior release isn’t the model architecture — it’s the access mechanism. For the first time, the U.S. government will vet who gets to use the frontier model, turning GPT-5.6 Sol into something closer to regulated infrastructure than a consumer product. The Washington Post reports that OpenAI agreed to a government-vetting framework for high-capability models, essentially outsourcing access decisions to a federal body. Sol Ultra, the top-end variant, will also land inside Codex for developers.
🎩 Cask’s Take
The three-tier naming — Sol, Terra, Luna — is the most telling signal here. OpenAI isn’t just releasing a better model; it’s structuring its product line like an aerospace company. Sol is the heavy lifter you need clearance for. Terra is the everyday workhorse for businesses. Luna is the edge runtime that fits in your pocket. This is the same playbook as cloud infrastructure (compute tiers, access tiers, pricing tiers), applied to intelligence itself. The government-vetting piece changes the game entirely — it means frontier capability is now treated as a dual-use export, even for domestic users. The era of “download and run” at the frontier is over. It’s now “apply, wait, and maybe get approved.”