ClaudeCode does it - Codex does it - Stop being shocked

Both of these big companies use a trick repeatedly with their vague seemingly amazing subscriptions. What you think is model nerfing is actually a highly automated system kicking in to engineer model usage AND HYPE !!! People think the business is in the model, it's not, it's in this system that I'm describing. SIMPLY PUT When there is a new model just released, the next one is not being trained as intensly, GPUs are free to run the latest model; the objective is to get people hyped and have users from the other company switch sides so all the resources are allocated to intensely running the latest model as efficiently as possible. But once the switching settles resources must be shifted toward training the next model. This is two fold: 1. train new model 2. make the difference between next and last model seem big It's not a knob either that someone decides to twist. This is definitely an intricate, complex, automated system with thousands of factors going in, the output of it is: what percentage of our money should go towards giving people LLM compute to MAXIMIZE revenue. That means minimize churn but also minimize cost too. Next time you're paying $200 for your subscription just whisper this: "There is no free lunch" so you won't be shocked when the glorious Opus 5.5 starts acting like a gemini flash lite 2.0 preview