The Price of Intelligence Is Falling Rapidly

An Epoch AI analysis by Emberson and Roodman documents a dramatic fall in the cost of achieving a fixed level of AI performance: about a 47% decline per quarter over the past three years, roughly a 13-fold reduction each year and a faster pace than any other transformative technology. The report gives a concrete benchmark: OpenAI’s o3 required about $0.30 per question to score 75% on GPQA Diamond in January 2025, while GPT-5.6 Luna reached the same score for about $0.0004 per question by mid-2026 - around a 725-fold cost reduction in under 18 months.

The key takeaway is that intelligence is becoming both smarter and far cheaper to run; the inference expenditure needed for a given performance level is plunging. That decline in operational cost matters as much as raw model capability because it makes high-performance models widely accessible and undermines the notion that only closed, frontier systems pose the main risk - frontier models are not only stronger but rapidly cheaper at any given level of performance, accelerating diffusion and changing the economics of AI deployment.