OpenAI dropped GPT-5.6 on June 26, 2026. Three models—Sol, Terra, Luna—each designed to own a different slice of the AI market. But the real headline isn't the benchmarks or the pricing tiers.
The real story is that the US government asked OpenAI to keep its most potent model locked inside a small group of vetted partners. And OpenAI said yes.
Let that sink in. The company that built its brand on democratizing artificial intelligence—the one that started as a non-profit promising to give away AGI—just voluntarily handed over the keys to a handful of government contractors and defense labs. The rest of us? We wait.
Sol is a beast. It sets new records on coding benchmarks, demonstrates autonomous vulnerability discovery in cybersecurity, and improves protein folding predictions. Capabilities that sound like science fiction are now real enough that national security agencies worry about them getting loose. That's why Sol's access is restricted, not because of safety research, but because of dual-use concerns—cybersecurity and biological research.
But here's the twist: OpenAI also launched Luna, a budget model at $1 per million input tokens, and Terra, a mid-range workhorse. The three-tier strategy is a masterstroke for market capture. Luna threatens every low-cost API provider. Terra squeezes Claude Sonnet and Gemini Pro. But Sol is the crown jewel, and only the well-connected can touch it.
Small startups and independent developers are the quiet losers. They built businesses on OpenAI's API, assuming equal access. Now they face a two-tier system where competitors with government ties get early crack at frontier capabilities. That's not a bug—it's the new feature of AI development.
The irony is deafening. OpenAI, once the poster child for open AI, now partners with the government to restrict access. Whether you call it prudent caution or a betrayal of founding principles, one thing is clear: the era of unconstrained AI releases is over.
Trends suggest we're heading toward 'AI feudalism'—where access to the most capable models depends on political connections rather than merit. If Sol remains locked for months, expect backlash from the research community. Congressional hearings might follow. And competitors like Anthropic and Google are reportedly working on their own tiered responses.
So who wins? The US government and its chosen partners. Who loses? Everyone else. And the question we should all be asking: If the most powerful AI is only for the powerful, what happened to the promise that AI would be for everyone?