AI Token Prices Drop 50 Percent as Demand Accelerates
Citadel Securities reports a significant decline in AI token pricing, driving a shift toward cheaper models and increased overall global demand for AI inference.
Average prices per token for artificial intelligence models fell significantly through August 2026, with consumption-weighted prices dropping approximately 50 percent over six months. Citadel Securities reported that average token prices at the frontier declined by 62 percent, a trend driven by competitive pressure and the rising market share of cheaper open-weight models.
This shift, described as Tokenomics 2.0, occurs alongside high costs for renting necessary compute chips. Data from Ramp indicates that businesses are responding to these price changes by routing routine work to cheaper models, with the usage share of top-tier models dropping from 53 percent to 45 percent in a single month this summer.
Analysts from Andreessen Horowitz note that this trend illustrates Jevons paradox, where lower unit costs lead to increased overall spending because the volume of usage rises. While the price decline may erode margins for frontier labs over time, the total volume of tokens suggests that global demand for AI inference is accelerating. To manage costs, business leaders are advised to price finished tasks rather than individual answers and optimize model effort settings provided by developers like OpenAI and Anthropic.
