Thanks for sharing. To me, purely on personal preference, the Gemini models did best on this task, which also fits with my personal experience using Googles models to summarize extensive, highly specialized text. Geminis 2.0 models do especially well on Needle in Haystack type tests in my experience.