VentureBeat Jan 26 Qwen3-Max Thinking beats Gemini 3 Pro and GPT-5.2 on Humanity’s Last Exam (with search)