OpenAI's Ted Sanders: Pick Astra for quality, Sol for balance, Terra or Luna for price

OpenAI's Ted Sanders has offered a rule of thumb for choosing between the company's GPT models: pick Astra for quality alone, Sol if you are somewhat price sensitive, and Terra or Luna if you are very price sensitive. The guidance, reported on October 27, 2026 by The Decoder (Matthias Bastian), accompanies Sanders' categorization of the new GPT-6.1 Terra.

Performance is multidimensional

According to Sanders, model performance has many dimensions. A model that excels at math will not automatically be good at jokes. Even so, he says most performance gaps come down to one variable: model size.

Sanders says OpenAI's model names are hard to simplify because model size, reasoning effort and multi-agent systems are distinct dimensions. Combining them into a single "IQ score" would obscure details.

Size does not explain everything

Sanders notes that size is not the whole story. Newer, smaller models can outperform older, larger ones through better training data or improved architectures.

Where GPT-6.1 Terra sits

Sanders says GPT-6.1 Terra is likely better than the larger Astra at technical communication. He says it roughly matches Astra in image processing and computer use at a much lower price point, but falls short in most other areas.

His summary guidance: "If you care about quality only, pick Astra. If you're a bit price sensitive, pick Sol. If you're very price sensitive, pick Terra or Luna."

What this means

The guidance treats model choice as a trade-off between quality and cost rather than a single ranking. Sanders' point that size, reasoning effort and multi-agent systems are separate dimensions explains why OpenAI's names do not reduce to one score. His comparison of Terra and Astra shows the practical consequence: a smaller, cheaper model can match or beat a larger one on specific tasks such as technical communication, image processing and computer use, while trailing in most other areas. For builders, the task profile and price sensitivity matter more than model size alone.