A number alone says little: is 90 tokens/s fast? Pick a model, add cards — and even other models — then watch the same answer being written at the speed of each combination.
Simulation: pre-written text, replayed at the speed measured in the database — or estimated when the combination has not been measured yet (column marked “estimated”). No LLM runs here.
Cards (1/4) · click = add a column for this model
green = measured · orange = estimated · switch models between additions to compare different models.