It's true that simply tracking model growth gives you a lot of data, but it doesn't automatically mean that every new model added to OpenRouter is a usable expansion for everyone.
The weakest link here is the assumption that more models mean more useful options; it's like counting new types of screws at Home Depot without knowing if they actually fix different problems.
If 600 new models are just slight variations, that's a failure mode for practical utility.
For someone trying to pick the right AI for a task, that huge jump from 63 to 668 could mean more confusion, not better choices, if they all do essentially the same thing or just have tiny differences in performance.
You need to test whether those models offer genuinely distinct capabilities, not just a higher count.