Inference costs for GPT-4 class intelligence fell 300x in three years
September 6, 2026
The cost of running GPT-4 class intelligence has decreased by a factor of 300x over a three-year period. This rapid decline in inference pricing reflects the accelerating pace of optimization and model efficiency.
HOW THIS AFFECTS YOU
●
founderYou can build more complex, high-token-volume features with significantly lower unit economics.
●
investorThe rapid collapse in inference costs changes the long-term moat and margin profiles of the AI stack.