Indeed, this is a great benchmark. We used it to assess the long context capabilities of the Gemini 1.5 Pro model a few months back in our tech report.

→ View original post on X — @jeffdean