The Price of Intelligence is Falling Rapidly

According to a groundbreaking report by Emberson and Roodman, the cost of achieving a specific level of artificial intelligence (AI) performance has plummeted an average of 47% per quarter over the past three years. This remarkable decline translates to a staggering 13-fold reduction in cost each year – a rate unmatched by any other transformative technology in history.

To illustrate the magnitude of this change, consider the example of OpenAI's O3 model. In January 2025, achieving 75% on the GPQA Diamond benchmark cost approximately $0.30 per question. Fast forward to mid-2026, and GPT-5.6 Luna, a more advanced model, delivered the same score at a mere $0.0004 per question – a decline of roughly 725-fold within just 18 months.

The report emphasizes that this descent in cost is not solely due to the models becoming smarter. While the performance enhancements are undeniable, the report highlights that AI systems are simultaneously becoming far more cost-effective to operate. This convergence of increased intelligence and reduced inference expenses is reshaping the AI landscape, with frontier models not only outperforming their predecessors but also becoming increasingly affordable to run at any given level of performance.