Quoting Ted Sanders
OpenAI's model tier order is Luna < Terra < Sol < Astra, driven primarily by size.
“In general, models that are bigger will (a) be more performant / reliable, (b) cost more per token, and (c) have better recall of esoteric facts.”
OpenAI's Ted Sanders explains on Hacker News that the company's model lineup follows a size-based hierarchy (Luna < Terra < Sol < Astra), with model size as the dominant factor explaining performance differences. He also clarifies that version numbers encode release order, family relationship, and improvement magnitude simultaneously. This is useful practitioner context but not a major announcement.