Without Caching, Your AI Agent Costs 10x More - George Cameron, Artificial Analysis

Some models now cost dollars to complete a single task, while others cost a few cents. George Cameron, co-founder of Artificial Analysis, breaks down what actually drives agent costs in 2026: input tokens, cache hit rate and the number of steps. In one example, an agent task costs about $21 with caching and over $200 without it. He then shows how to decide when you need the smartest model and when a cheaper one gives the same answer. 0:00 Intro to Artificial Analysis 2:47 The AI landscape and the Intelligence Index 5:25 Progress is accelerating 6:27 Open weights models stay close behind 6:53 Why cost matters now 8:53 Intelligence gets about 10x cheaper every year 10:18 Breaking down the cost of an agent task 12:59 Why caching matters so much 13:38 Reasoning effort and cost 15:30 Making the trade-off decision 17:02 Limited vs. unlimited intelligence tasks 18:12 How to shortlist and benchmark models George Cameron, Artificial Analysis: https://artificialanalysis.ai Join us in SF for the second annual AIE CODE summit: https://aie.codes our last flagship summit of 2026. Applications live now!