We've reached the model abstraction era a lot sooner, which means the significance & superiority of each frontier model will only go down from this point. A cluster of small models can outperform the frontier at cost efficiency, speed & capability. That implies that scaling laws & aggregation theory have already commoditized frontier models. What seems like model routers will quickly morph into model aggregators that both control the relationship with the end users & own the data of each models' performance to determine which was the optimal choice. That's inefficient at present, but after accumulating enough data to index the performance per model, the cache needed & tokens used to optimize it automatically per task.