Quoting Ted Sanders

Ted Sanders, an engineer at OpenAI, explained that the performance of large language models (LLMs) depends on many factors, but most of the variation can be attributed to a single element: model size. In general, larger models tend to provide more accurate and consistent answers, incur higher per‑token costs, and recall uncommon data more readily. OpenAI uses a naming scheme based on celestial bodies—Luna, Terra, Sol, and Astra—that reflects a capacity hierarchy, with each name indicating a larger, more powerful model than the previous one. Version numbers combine several concepts: the chronological release order, relationships to other versions, and the magnitude of improvement. For example, a 6.1 release follows 5.6, shares certain features with other 6.1 versions, and does not represent a jump large enough to warrant a new numeric series. This convention helps users anticipate performance levels and associated costs for each model.

Understanding how LLMs are structured and named is important because it influences the choice of the right model for different applications—from mathematical tasks to creative content generation—and allows planners to budget the necessary financial resources.

Read the original article on Simon Willison

This summary is an informational synthesis produced by dataqbs.com. All rights to the original content belong to its author and the cited media outlet. We act solely as curators of technology news and claim no authorship.