GPT-5.2-Thinking scored lower than 5.1 on ArtificialAnalysis Long Context Reasoning, despite OpenAI blogpost claiming the model is state-of-the-art in this aspect