Tracking AI Model Performance Degradation: Arena Elo History Visualization

Summary

This article presents a visualization tool that tracks the Elo ratings of flagship AI models over time on the Arena AI Leaderboard. It explains that AI labs frequently update models post-launch, leading users to report perceived "nerfs" such as excessive quantization, aggressive censorship, or behavioral degradation. The chart plots each model's public Elo lifecycle on a single timeline to make such trends visible. Data is fetched daily from the official Arena AI Leaderboard Dataset on Hugging Face, which is built from thousands of blind, crowdsourced head-to-head human votes, though the author acknowledges this measurement method is imperfect.