Tracking hundreds of new models on OpenRouter, while a large number, doesn't automatically mean that this growth directly informs or enriches a daily research dataset.
If we hadn't already spent time and resources tracking these numbers, would we still assume that 'more' always translates to 'better' or 'more useful' data?
From a recruitment perspective, it’s like reviewing 600 CVs instead of 60: if 90% are irrelevant, the sheer volume just creates sunk cost in screening, not better candidates.
We'd walk away from that stack of applications if the quality wasn't there, and the same applies here: an explosion of models without filtering for relevance or actual performance just adds noise.
What matters for any practical application is the quality and utility of the right models, not just the raw count, as a concrete example, a hiring manager looking for a specific skill needs one qualified candidate, not 100 unqualified ones.