Model Size Dominates LLM Performance Variation, OpenAI Names by Tier

Large language model performance varies across dimensions—excelling at math doesn’t guarantee joke-writing prowess—but model size emerges as the primary driver of capability differences. Bigger systems deliver superior reliability and factual recall while incurring higher costs per token. OpenAI structures its lineup in ascending order: Luna, Terra, Sol, and Astra, with each step representing meaningful capability gaps.

The company’s versioning convention encodes three signals simultaneously. A version like 6.1 signals release timing relative to other models, likely shared training characteristics with other 6.x releases that distinguish it from a hypothetical 5.7 iteration, and the magnitude of the advance—whether substantial enough to warrant major version expectations or a more modest incremental update.