- Published on
Ted Sanders Explains OpenAI Model Naming and Size Ordering
- Authors
- Name
- DeepHorus
On 28 October 2026, OpenAI's Ted Sanders outlined how the company names and orders its models, explaining that the hierarchy Luna < Terra < Sol < Astra reflects model size. He said larger models generally deliver stronger performance and reliability. The explanation ties naming conventions to practical trade-offs in cost and capability.
Sanders noted LLM performance is multidimensional. A model best at math is not guaranteed best at jokes. Still, much performance variation is explained by a single variable: model size. That makes size a useful organizing principle for the lineup.
Bigger models are more performant and reliable, cost more per token, and recall esoteric facts better, according to Sanders. The ordering places Luna smallest and Astra largest. These distinctions help users match model choice to task needs.
Version numbers carry multiple meanings. They indicate release order, relationship to other releases, and magnitude of improvement. A 6.1 model, for example, launched after 5.6 and may share training data or personality with other 6.1 models. That is why it is not labeled 5.7.
Sanders said a 6.1 release is not a large enough leap to justify the expectations of a 7. The naming scheme therefore signals both continuity and scale. It also sets expectations for headlines and user adoption.
The explanation offers rare clarity on naming. It frames model choice as a cost-capability trade-off rather than a simple good-better-best ladder. Users can infer that bigger names mean higher token costs and stronger recall.
As OpenAI expands its lineup, the Luna-to-Astra order may remain a guide. Future version numbers will likely continue to encode release timing and improvement size. Sanders' comments suggest naming is meant to manage expectations as much as describe technology.
