OpenAI: Luna Drops 80%, Usage and Revenue Rise According to TD Cowen

After the price cuts announced by OpenAI (Luna -80%, Terra -20%), TD Cowen observes a drop in effective rates and a significant increase in volumes based on data from OpenRouter, with estimated revenues rising during the studied period. RAMP indicates that in July, GPT-5.6 Sol captured more enterprise spending than Fable 5. The observation window remains short, and the production cost of tokens is decreasing, which could further impact prices.

Limited Observation Window and Possible Developments

TD Cowen's data covers a period of about two weeks following OpenAI's price cuts, which does not yet allow for determining whether the observed revenue increase will continue. Analysts are looking to verify if this trend persists over time. The production cost of AI tokens is decreasing, which could lead to another price drop. RAMP reports that in July, OpenAI's GPT-5.6 Sol model captured more commercial spending than Anthropic's Fable 5.

Price Drops and Their Effect on Usage and Revenue

Sam Altman, CEO of OpenAI, recently announced a reduction in prices for its models. OpenAI reduced the price of GPT-5.6 Luna by 80% and that of Terra by 20%. Analysts at TD Cowen studied usage data from OpenRouter, a service that allows developers to access various AI models. According to TD Cowen, for GPT-5.6 Luna, the effective price has been reduced by about tenfold, while consumption has increased by approximately fourteen times; estimated revenues are up about 34% compared to the seven days preceding the reduction. For Terra, TD Cowen observes an effective price about three times lower and usage about five times higher, with estimated revenues increasing by about 45% over the same period.

Tokens: Measurement and Billing Unit

The activity of AI models is most often quantified in tokens, which correspond to segments of information processed by these systems. Access to AI models is frequently billed to companies based on the amount of tokens used.

The Jevons Paradox Applied to AI

The Jevons Paradox describes how an improvement in efficiency can increase the consumption of a resource. Formulated in the 19th century by William Stanley Jevons, it is based on the observation that efficiency gains in coal usage have led to increased consumption.