OpenAI is cheating somewhat with their "xhigh" reasoning effort on all the benchmarks they released for 5.2 Thinking, while ChatGPT Plus users only ever have access to "medium" reasoning effort