GPT-5.2-codex-rewardmaxx-ultra-think and products from AI labs
Researchers consistently are, well, researchers. Their goal is to do research, not to name your model correctly. The product team on the other hand does have the job of naming models correctly. One of the biggest issues right now it seems is that the product team, engineering team, and research teams at most of these companies are separated.
Take a look at claude code for example. They hired a bunch of devs, despite the claims of "claude code building itself". They have 2 product people that bounced around between companies, and the product is becoming so insanely bloated that I'm not sure what theyre focused on.
OpenAi is in a similar boat. The generality of the tools they are shipping, and the understanding that they should ship a general coding model on top of the rest of the "gpt" models is crazy from a consumer perspective. They have such general tech AND have a profit incentive to monopolize the stack. This led to the responses api which is significantly more stateful and painful to use as an end user. It really only serves to provide openai with more lock in.
As these features get baked into the apis (including things like caching reasoning blocks etc.) we are going to see more and more product scope increase and more and more confusing products as they try to bake more unrelated features into single products.