The Economics of AI Model Pricing: A Quality-Adjusted Analysis of Commercial AI Models

Abstract

As the commercial AI landscape rapidly evolves, companies, governments, and academic institutions are investing increasing amounts of capital into artificial intelligence. However, the relationship between AI model cost and performance is more complex than previously assumed, making it difficult for users to identify cost-effective models. We examined how model characteristics, such as capability, openness, and pricing structure, influence AI model prices after accounting for differences in performance. Using data from 42 AI models across 11 laboratories between 2023 and 2026, we constructed a quality-adjusted hedonic price index using log-linear regression with laboratory-clustered standard errors to estimate the effects of model capability and business model on pricing. Our results show that while average AI model prices increased by 1.5× per year, the price of the cheapest available models declined by 22.7× per year. We also found that open-weight models were 72.5% less expensive than comparable closed models, with capability explaining 34% of this price difference. These findings suggest that quality-adjusted pricing provides a more accurate measure of AI market trends than raw prices alone and can help businesses and policymakers make better-informed decisions about AI investment and adoption.