I’ve spent the last few weeks immersed in reasoning and deep research models. I’ve seen enough. At least one (if not multiple) models will cross 50% on the Humanity’s Last Exam benchmark by th...