OpenAI’s steep AI price cuts are putting a high-stakes proposition to the test: sell each unit for less, and customers may buy so much more that revenue rises anyway.

The experiment began with an 80% cut to GPT-5.6 Luna and a 20% reduction for the mid-range Terra model. According to TD Cowen’s analysis of OpenRouter usage, Luna’s effective price dropped roughly tenfold while consumption jumped about 14-fold. Terra became about three times cheaper, yet usage rose fivefold.

That imbalance matters. The analysts estimated Luna revenue rose about 34% in the seven days after the reduction compared with the previous week, while Terra revenue increased about 45%. It is an early, unusually stark example of the idea that cheaper computing can unlock new demand rather than simply erode sales.

The bullish interpretation is that lower token costs make AI viable for work that once failed the economics test: processing more documents, handling customer queries, generating software and powering multi-step automated agents. The source report argues that the pattern resembles Jevons paradox, in which greater efficiency expands total consumption. “The cheaper AI gets, the more of it we use.”¹

But the evidence is preliminary. The TD Cowen figures cover only about two weeks after the cuts, and the analysts said the industry still needs to see whether the revenue bounce lasts. Price reductions may be drawing a burst of experimentation rather than establishing a lasting demand curve.

The timing also sharpens the competitive stakes. OpenAI is battling Anthropic for business spending while Chinese AI rivals add pressure to make capable models cheaper. Ramp found that OpenAI’s GPT-5.6 Sol captured more enterprise spending in July than Anthropic’s leading Fable 5 model, a result attributed in part to OpenAI’s lower price.¹

As computing systems make tokens cheaper to produce, the price war is unlikely to stop. The question is whether OpenAI has found a scalable revenue engine—or merely accelerated the race to the bottom.