Comment by mikejulietbravo

The short answer is that it depends how well you define the boundaries of the task and the relative complexity. For example, smaller model is usually fine for something like summarization, but an "easier" coding task might still actually be quite difficult unless you eval it heavily like @paulhoule said