Familiar! The Artificial Analysis Index is the metric models are sorted by in my sheet. But their data and presentation has some gaps. I made this sheet to get

Familiar! The Artificial Analysis Index is the metric models are sorted by in my sheet. But their data and presentation has some gaps. I made this sheet to get a glanceable landscape view comparing the three key dimensions I care about, and fill in the missing evals. AA only lists scores for a few increasingly-dated and problematic evals benchmarks. Not just my opinion, none of their listed metrics are in HuggingFace Leaderboard 2 (June 2024). That said I love the AA Index score because it provides a single normalized score that blends vibe-check qual (chatbot elo) with widely reported quant (MMLU, MT Bench). I wish it composed more contemporary evals, but don't have the rigor/attention to make my own score and am not aware of a better substitute.