The thing that bothers me is the false advertising. The models are announced with specific capabilities and benchmark scores. Then they silently nerf the models after release. Without a public acknowledgement that a production model has been nerfed and what the new benchmarks are, people are assuming the model has the original capabilities. People are choosing to pay for models based on their reported capabilities. Customers trust these models to deliver the reported capability. It feels like false advertising to me when they say one thing and deliver something less silently hoping no one finds out.

If I order a 16oz beer and get a 12oz pour, I'm going to call it out.