A quote from Ted Sanders

Why it matters

Ted Sanders of OpenAI has offered a rare look at how the company names and orders its models, addressing a persistent source of confusion for users trying to compare releases. He explains that OpenAI ranks its models by size—Luna, then Terra, then Sol, then Astra—with larger models generally delivering stronger, more reliable performance, charging more per token, and recalling obscure facts better. He acknowledges that language model performance is highly multidimensional, so no single model dominates every task, yet size alone explains much of the variation users observe. Sanders also lays out the logic behind version numbers: they signal release order, how a model relates to other releases, and how large an improvement it represents. A point release launched after a predecessor may share training data or personality with sibling releases, while a full version jump is reserved for advances big enough to justify the attention a major release attracts. The clarification gives readers a framework for interpreting future OpenAI announcements and pricing tiers, and it comes directly from the company rather than outside speculation.