I don't understand why one would let a router handle their LLM request (other than for routing it to the cheapest inference provider for open models).

When I build an application that uses LLMs, it's tuned and tested for a specific model only. LLMs are not really drop-in replacements of each other, even different versions of the same series.

The pertinent question isn't why OpenRouter but why a router company is worth 7B.

You can use one key for multiple models, you can have many keys with a budget on each, you can manage all your keys in one piece of glass, you don't have to deal with the upstream vendors directly just to name a few. These are all things that may be important to developers. I can't even get an OpenAI API key because those morons can't get their phone verification to work properly. OpenRouter is the only way I can use their models.