A LocalLLaMA community post proposes a benchmark that would score models as “good enough” when they meet a 95% confidence threshold, with results considered alongside costs.
The post asks whether there is a benchmark for people without specialist expertise. It proposes awarding 100% to any model that reaches a 95% threshold, then comparing models based on whether they are good enough for everyday users.
The author suggests relating benchmark results to costs, including both on-premises and cloud use. The post does not describe an existing benchmark or report test results.