LLMs: Intelligence vs. Cost

ArtificialAnalysis’s Intelligence Index ranks LLMs by benchmark performance and reports the cost of running each model, plotting a Pareto frontier of cheapest models for given intelligence scores. The author criticizes the original plot’s logarithmic cost scale, reliance on official API pricing, and inclusion of datacenter prices for locally runnable models, then presents revised linear-scale charts that incorporate third-party rates and electricity-based costs for on-device models, highlighting diminishing returns between intelligence and expense.