OpenAI is cutting the price of its models hard—and, at least in the first burst of data, customers are buying so much more AI that the discounts may be paying for themselves. The catch is that the company is making that wager in an increasingly brutal global price war.

The shift began with sharp reductions for two OpenAI models. GPT-5.6 Luna was cut by 80%, while the mid-range Terra received a 20% reduction. TD Cowen’s analysis of OpenRouter data found Luna’s effective price fell roughly tenfold after the changes, yet consumption climbed about 14-fold. Terra’s price dropped by about threefold and use rose fivefold.¹

That mattered because the extra demand appeared to outrun the lower per-token price. Over the seven days following the cuts, TD Cowen estimated Luna revenue rose about 34% and Terra revenue about 45%. “The remarkable part is that usage rose faster than prices fell,” the analysis said—a result that turns a conventional margin squeeze into a volume-led growth play.¹

The emerging logic is straightforward: cheaper models make previously marginal tasks worth automating. Businesses can process more documents, handle more customer queries, generate more code and run multi-step agents at a cost that was once prohibitive. OpenAI’s GPT-5.6 Sol also captured more business spending in July than Anthropic’s top Fable 5 model, according to Ramp data cited in the analysis, with price presented as a likely advantage.¹

But the early rebound is not proof of a durable win. The data covers only about two weeks, and analysts said they need to see whether the pattern persists. Meanwhile, the competitive frame is widening: OpenAI and Anthropic are in a price war as Chinese AI rivals gain ground.² Lower inference costs could keep pushing prices down—forcing every major model maker to prove that exploding usage can offset shrinking unit economics.