Each mark represents a model’s median score across eligible public attempts. Models appear after at least 5 public attempts. Cost is billed USD rather than list price. Marks are tagged frontier, standard, flash, or open; use search and the controls to filter by name, lab, or tier. The score control can switch the plotted score from overall to a subject median (coding, logic, perception, research, data analysis). When score is on the vertical axis and cost/time/tokens on the horizontal, a pareto line traces undominated models (cheaper-or-equal and higher-or-equal score). Click a mark to compare models, or click a lab in the legend to filter the view.