Ted Sanders on picking between Luna, Terra, Sol and Astra
OpenAI's Ted Sanders on choosing between Luna, Terra, Sol and Astra: pick by price sensitivity, with Terra for clear writing and fast computer use.
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.
Ted Sanders, who works at OpenAI, in a Hacker News comment on the GPT-6.1 Terra launch thread (October 26, 2026)
Context: someone in the thread asked why OpenAI ships four model sizes. Sanders opened with “I work at OpenAI and I can try to explain our logic,” gave the order as “Luna < Terra < Sol < Astra”, and said “much of the variation in performance is explained by just a single variable: model size.” On Terra specifically, he wrote that it’s “probably better than Astra and Sol at technical communication, similar on vision and computer use (but much faster!), and a bit worse on most other things.” Simon Willison quoted that part. I priced what “very price sensitive” buys you on an agent session in GPT-6.1 Terra: what an agent session really costs.
— Ted Sanders