Not even an average task. I can have a single task that I need to do and I could be choosing which model to use. The cheapest-per-benchmark-task model would be useless to me if it cannot do the task I need.
I'm sure there are degenerate cases, but I'd bet a relatively small model could do the job.