OpenRouter may put too much weight on input prices when routing AI requests

A user claims OpenRouter routes requests by the inverse square of provider pricing: at $1/M versus $2/M, they say the cheaper provider gets four times the traffic. In a GLM 5.3 experiment comparing Inference.net and Novita, they say 70% of requests went to Inference.net. They suspect OpenRouter weighs input prices too heavily and say the result corresponds to a 95/5 weighting of input versus output prices.