The two cheaper models in the GPT-6 family came out on 22 September. Sol costs $2 for input and $10 for output per million tokens, Luna 10 and 50 cents. Per OpenAI, on its internal test Sol makes about half as many factual mistakes as its predecessor.
- Sol drops from $4 to $2 for input and from $20 to $10 for output; Luna from $0.20 to $0.10 and from $1.20 to $0.50.
- On OpenAI's internal factuality test, Sol makes about half the mistakes of GPT-5.6 Sol.
- The models are in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu, and free users get Luna in the desktop app.
In two days three labs released new models, and all three talk about price first.
OpenAI comes in with its two smaller models and the simplest argument on the market: half price. Astra stays on top for the heaviest work, and Sol and Luna are for everything else that runs all day.
The mistakes
The benchmark points are OpenAI's, measured at different effort levels, while the other models' numbers are taken from public reports.
The number that matters is the mistakes. The test is built precisely on conversations where people caught the model in an error, which makes it harder than typical use. Half the mistakes there means a lookup you will correct less often. It does not mean a lookup you can skip checking.
With Luna the output price falls more than the input price, from $1.20 to $0.50, and in a product with long answers the difference shows in the first month.