- Published on
Quoting Ted Sanders
- Authors
- Name
- DeepHorus
OpenAI's Ted Sanders has explained how the company names its AI models, clarifying that the lineup runs from Luna to Terra to Sol to Astra. He said the ordering reflects one dominant variable: model size. Larger models are generally more capable and reliable, cost more per token, and recall esoteric facts more accurately.
The explanation, shared on 28 October 2026, addresses confusion over OpenAI's use of names and version numbers. Sanders noted that LLM performance is multidimensional, so no fixed order guarantees that the best model at math is also best at jokes. Still, he said much variation is explained by model size.
Under the hierarchy, Luna is the smallest model, followed by Terra, Sol, and Astra. The progression implies increasing performance, higher per-token costs, and stronger recall of obscure information. The naming gives users a shorthand for capability and cost before they compare benchmarks.
Sanders also detailed what version numbers mean, saying they signal release order, relationships to other releases, and magnitude of improvement. A model labeled 6.1, for example, launched after 5.6 and may share training data or personality traits with other 6.1 models. That similarity explains why it is not called 5.7, he said.
He added that a version like 6.1 does not represent a large enough leap to justify the attention and expectations of a 7. The distinction matters for customers choosing models for tasks from mathematical reasoning to creative writing. It also sets expectations for cost, reliability, and factual recall.
The disclosure offers a rare look at how OpenAI communicates model progress as AI competition intensifies. Sanders's comments suggest the naming system balances transparency with marketing restraint. Future names and version numbers will likely continue signaling relative size, release timing, and scale of improvement.
