It would be nice if the interface for reading the benchmark was more interactive. Sorting by different options, such as accuracy/$, accuracy/time (both of which are already in the benchmark, but you cannot sort based on them).
It could also be convenient to select models for custom Assistants based on these factors/results; e.g. a new model gets added that is more accurate or faster than the current best, it gets picked automatically if you select "best in accuracy" or "best in speed". At least a more accessible spot for the benchmark results would be nice – hiding it in a help page isn't very good UX in my opinion.
I'll also link to #6896, it would be useful to be able to see which models are available under the user's current plan, and filter out the Ultimate plan models so they can compare only those available to them.
