OpenAI Names Models Sol and Luna Because Subtlety Is Bankrupt

OpenAI has announced GPT-6 Sol and Luna, the latest additions to its expanding roster of large language models, each priced at roughly half the cost of their GPT-5.6 predecessors. The nomenclature tells you everything you need to know about how the company sees itself.

Sol—Latin for the sun—handles complex coding and agentic workflows. Luna, the moon, optimizes for efficiency in focused, high-volume tasks. Both were trained using similar methods as GPT-6 Astra, the heavyweight that Greg Brockman, OpenAI's president, recently described as the first tangible sign of "the AGI era." Three models, three celestial bodies, three escalating declarations of cosmic importance. It is confidence theater at scale.

Consider the framing: you don't name your products after the sun and moon unless you believe you're operating at a civilizational inflection point. You don't invoke the solar system casually. You invoke it when you've decided that the ordinary taxonomy of software versioning—the incremental, humble parlance of v2.1 or Enterprise Plus—is beneath the magnitude of what you're building. When you believe you're reshaping human civilization, why bother with restraint?

This matters beyond branding aesthetics. The naming strategy is a window into institutional psychology. It reveals how OpenAI's leadership thinks about its role in the world, and more importantly, how it wants everyone else to think about it. The celestial nomenclature says: we are not merely incremental. We are foundational. We are forces of nature.

Technically, the models do deliver on some promises. Sol and Luna cost half as much as GPT-5.6, which debuted in July following a staggered federal government-overseen release that itself reads like theater—the performance of regulatory oversight in an industry that moves faster than the bureaucracy watching it. The cost reduction matters for deployment. The efficiency gains matter for practitioners. But the naming? The naming is pure signal.

There's another subtext worth excavating. OpenAI trained Sol and Luna using methods similar to GPT-6 Astra, potentially including recurrent depth—a technique that enhances performance while making it harder for external researchers to monitor decision-making processes. This is the quiet part that doesn't make the press release. The celestial branding obscures a technical choice that raises legitimate safety questions. When you're naming your models after astronomical bodies, who's paying attention to the footnotes about interpretability?

The timing cuts deeper. Anthropic released a new version of Opus 5.5 approximately 90 minutes before OpenAI's announcement, underscoring the velocity at which frontier AI labs are now operating. This is not competition in any traditional sense. This is a sprint to some undefined finish line, conducted in public, with celestial metaphors replacing rigorous explanation of what these models actually do and how they should be used.

OpenAI has positioned Sol and Luna as solutions to "rogue agent behavior"—a category of risks that has animated AI safety debates in recent months. The irony is thick: you announce your concern about agents behaving unpredictably by deploying a product trained with methods that make behavior less transparent, then wrap the whole thing in sun-and-moon mythology that suggests cosmic inevitability rather than engineered choice.

This is not to say the models aren't genuinely capable. They likely are. But capability divorced from humility is its own form of risk. The celestial branding isn't incidental. It's declarative. It announces that OpenAI has moved past the stage of building tools and into the stage of building mythology. The company is no longer in the business of incremental improvement. It is in the business of historical inevitability.

The question is whether the market—and more importantly, regulators—will read the subtext embedded in the nomenclature. The sun and moon have no off switch. They have no oversight committee. They simply exist, indifferent to human concerns. OpenAI's models, whatever they're called, do not operate in that category. Yet. But the names suggest that someone at the company believes they're on a trajectory toward it.