OpenAI CEO Sam Altman has outlined a strategy to increase the quality of AI models while simultaneously reducing their costs. The company aims to deliver the most efficient and affordable solutions in the industry by focusing on a metric called Useful Intelligence per Dollar.
On July 30, 2026, the company reduced API pricing for the GPT-5.6 model family. The cost for the Luna model was cut by 80%, and the Terra model by 20%. Luna now costs $0.20 per million input tokens and $1.20 per million output tokens, while Terra costs $2 and $12 respectively.
Following these price cuts, Luna saw a 14-fold increase in consumption and a 10-fold drop in effective price, resulting in a 34% rise in revenue for the model. In September 2026, OpenAI further reduced prices for GPT-6 variants to half the cost of the previous series, citing improvements in caching and inference.
The company's pricing adjustments follow competitive pressure from models such as Kimi K3. By prioritizing high-volume usage of cheaper models, OpenAI is leveraging its infrastructure to drive revenue growth despite the high costs associated with computing power.
