GPT-5.6 price and performance reshuffle: Luna price reduced by 80%, single task cost approaches 6 cents
OpenAI pushed the price and performance of the GPT-5.6 series to a new frontier on July 30: Luna’s price was reduced by 80%, Terra’s price was reduced by 20%, and Sol added the fastest 2.5 times Fast mode; Luna provides cutting-edge performance a year ago at a single task cost of about 6 cents, and introduces end-to-end cost optimization of kernel self-optimization.
When the competition for large models shifted from "parameter competition" to "unit intelligence cost", OpenAI behind ChatGPT chose to answer the market with a complete repricing: On July 30, the three-speed models of the GPT-5.6 series were adjusted at the same time. The fastest and most economical Luna directly reduced its price by 80%, the balanced Terra's price dropped by 20%, and the flagship Sol received the highest price 2.5x accelerated Fast mode. This is not a promotion, but turning "cost-effectiveness" itself into a signal of product strategy.
Repricing of three-tier architecture
The division logic of the GPT-5.6 family essentially indicates to developers that "different tasks should pay for different capabilities."
| Model | Positioning | Input price (/million tokens) | Output price (/million tokens) | This change |
|---|
Reviews