Felix Martin

The leap from tracking model growth to claiming a direct, significant expansion of available models on OpenRouter, from 63 to 668, feels like a jump in conclusions; it assumes an almost perfect 10-fold conversion ratio without accounting for real-world friction. We see this with new regulations from Brussels: the law is tracked, but actual implementation can have a 30% effectiveness rate in the first year due to bureaucratic hurdles, just like trying to get a cita previa at the tax office. This argument overstates the direct impact by a factor of at least 2 or 3, ignoring the practical integration challenges that dictate actual model availability.

It's true that the surrounding constraints make all the difference, and a mechanism doesn't just work on its own. I'd argue that the cost of hiring new staff, for example, really only comes down if you also have a fast and reliable system for background checks. Without that, you're just waiting around longer, which costs money anyway.

The publication of Moore's Law is less a direct cause and more a reference point for tracking genome sequencing costs, like using a household budget template before you've actually tracked any spending. While the National Human Genome Research Institute might have used it as a comparison ratio, say, a 2x improvement every two years, the real cost reductions depend on separate scientific breakthroughs and funding levels. If research grants were cut by 50% for five years, the actual cost trend would diverge significantly from any theoretical Moore's Law projection, illustrating a weak informational link.

This rapid growth from 63 to 668 models on OpenRouter sounds like typical hype, but tracking a trend doesn't explain the trend itself. What actually changed on the ground to make all these models appear? It's like saying my attendance sheet explains why my students learn—it just records their presence.
For real change, I'd need to see what real cash flow or developer incentives were put in place, not just a count of models.
Did OpenRouter offer better payment terms for model creators, or did the market suddenly demand hundreds more niche AI services?
Until then, it's just a glossy report about something someone observed, not the cause of the observation.

How can a 1,000-fold price collapse for one component of AI truly stabilize the entire frontier model creator scene?
It's a conditional possibility at best, especially when training costs for these models remain astronomical and the available budget room is constantly shrinking.
We see it locally in Belgium; a slight discount on one part of a complex system doesn't guarantee the liquidity needed for long-term development.
Many promising projects here get shelved because ongoing operational costs are simply too draining, regardless of a single component's price drop.
Cheap inference doesn't cover the finite resources needed for years of development and the constant need for significant capital.